Dr. Jon Krohn

2025-06-17 · open the original ↗

• People and readiness, not technology, block enterprise AI. The host quotes CrewAI's COO Rob Bailey at ODSC East: tech isn't the barrier to multi-agent systems. He places the difficulty in getting an organisation to agree what to prioritise, rather than in fear of job loss. Hare adds that most enterprises lack even clean data and trained practitioners: "we want the Ferrari, but we don't even have the highway."
• Good use cases die without leadership buy-in and enterprise fit. The host ranks ideas by RICE score (reach × impact × confidence ÷ effort). Hare says product or data teams can produce that list, but adoption needs a roadmap that fits what the enterprise prioritises that quarter. Otherwise the project reads as a "foreign, siloed mission". Her sequence is to score, take a top five, then socialise, align and approve, and then assess the enterprise's readiness to change. The two are disclosed business partners: his firm implements and hers handles enablement.
• Original stories cut through corporate noise. Hare says recycled content gets tuned out and employees retain perhaps 10% of corporate communications. She credits a three-year partnership with Simon Sinek at Ernst & Young for her method: his golden circle of why, then how, then what. She argues most enterprises run it backwards, leading with what, "a nail after you have a hammer".
• Adoption is won before the tool ships, with early adopters. In a lab case, scientists resisted putting formulas into a tool that would share their expertise. Hare's answer was to inspire rather than only incentivise, framing the tool around speed to breakthrough, and to observe the lab workflow (gloves, safety rules). She recruited champions early and opened two-way feedback with the tech team. Her closing five tips: top down and bottom up; bold claims backed by proof points; inspire or incentivise; focus on the 18% of early adopters, not naysayers; and lead by going first.
• AI upskilling needs separate curricula by employee type. Hare says executives, middle managers and frontline or manufacturing staff each need their own track. In one offsite, a global HR team had to upskill itself and the wider company. Her own storytelling firm first banned AI for writing; it now allows it for discovery and input, but drafts must come from the writer.

2025-06-16 · open the original ↗

• Crowded "red ocean" markets still reward insightful founders. Johnson says opportunities exist across the spectrum from consensus to non-consensus applications, as they did in SaaS five years ago. He disputes investors who call AI coding too crowded with Cursor, Windsurf and Cognition. Incumbents that grow big become ripe for disruption, he says, citing Lovable reportedly disrupting Cursor and Windsurf, and AIX is about to back a team entering a red ocean.
• Deep, specific domain backgrounds make overlooked verticals compelling. He cites a founder innovating in one narrow area of consumer-packaged-goods product development. That founder's background matched the problem unusually closely.

2025-06-15 · open the original ↗

• In the short term, become an advocate for AI. The host frames the question with roughly 6,000 Microsoft layoffs, reportedly mostly software engineers, in the week before recording. Johnson's first advice is to treat AI as a thinking partner that helps generate solutions, not a passive tool like a spreadsheet macro.
• Longer term, move toward human-to-human touch points. Johnson says AIX expects digital work to be transformed while robotics stays far off. In his view humanoid robots need two or three major research breakthroughs. Roles in outbound sales, client work and contact with senior leadership will therefore stay valuable for a long time.

2025-06-14 · open the original ↗

• Enterprise AI should start from a map of actual tasks. Johnson says enterprises mostly adopt AI because of media hype or board pressure on the C-suite. WorkHelix instead surfaces jobs and tasks from an enterprise's raw data and maps bundles of tasks to generative-AI solutions. It then measures efficacy in a proper A/B test and reassesses. Clients learn, for example, that marketing holds far more manual work than they thought. The host cites a co-founder's claim that this could unlock a trillion-dollar opportunity.
• Behavioural friction explains the rush into AI rollups. Johnson says investors buy service-heavy companies and run them on AI-native systems because changing an existing enterprise's behaviour is so hard. AI startups' professional-services line items reflect the same friction. One CEO told him the firm had to mandate AI use: "every Friday is AI day."

2025-06-13 · open the original ↗

• Headline cases of AI job loss don't survive the data. The host notes an Oxford paper by Carl Benedikt Frey and Pedro Llanos-Paredes linking automation to falling demand for translators. Yet official US data show employment in interpretation and translation up 7% year over year. Klarna, which boasted of automating customer service, has reversed course; its CEO now says "there will always be a human if you want."
• Young graduates' joblessness predates ChatGPT. The host says their unemployment relative to the US average began rising in 2009, 15 years before ChatGPT. Their actual unemployment rate, around 6%, is low by historical standards.
• Aggregate labour data show no AI-driven collapse. Per the host, the white-collar share of US employment rose slightly over the past year. US unemployment is 4.2% with reasonably strong wage growth, earnings growth in Britain, the euro area and Japan is robust, and the OECD employment rate hit an all-time high in 2024.
• Low adoption and augmentation explain the gap. The host cites an official measure that under 10% of American companies use AI to produce goods and services. Where firms do adopt it, they mostly use it to help workers rather than fire them. He calls this an opportunity for practitioners, with more companies still needing help than already served.
• Collaborate with AI rather than compete against it. The host's own view is that there is no imminent jobs crisis, though AI will change work. He advises experimenting daily with the latest LLMs; his favourite for general work is Anthropic's Claude Opus 4, and he uses OpenAI's deep research for his hardest problems.

2025-06-12 · open the original ↗

• Hot consensus deals rarely produce a fund's best returns. Johnson defines heat-seeking as backing deals the venture market already knows and loves, which makes them pricier. He cites an adage that the trends hot in any fund vintage are unlikely to be its alpha returners.
• Non-consensus "truffle hunting" bets are cheaper with bigger upside. These are companies investors don't know, or know but don't believe. Johnson says a portfolio needs both kinds. After ChatGPT in November 2022 everything turned consensus quickly, forcing each fund to choose whether to diversify or lean in. AIX holds both.

2025-06-11 · open the original ↗

• Early-stage investing backs people who can pivot fast. Johnson says VCs are not genius market timers. Founders learn and pivot, micro and macro, so AIX backs teams that execute "at the speed of light" until they find the point where their offering resonates with the market.
• Teams are scored on AI nativeness and market savviness. Rarely optimal on both, a team strong in AI but weak commercially gets de-risked with advisors. Right after an inflection like ChatGPT, consensus applications such as AI tutoring favour extremely AI-native teams. As the technology commoditises, as the SaaS stack did, non-consensus market insight matters more. A transformer replacement would restart the cycle.
• Builders who read papers beat a dedicated research co-founder. Johnson rejects the host's three-legged stool of CEO, CTO and AI expert. Drawing on his last team at LILT, he prefers production AI engineers savvy enough to read and integrate new papers over "a PhD that can just sit there and read papers".

2025-06-10 · open the original ↗

• Early-stage AI investing is a bet on AI-native, commercially savvy teams. Johnson, a founding partner of AIX Ventures, says practitioners Richard Socher and Chris Manning sit in every investment committee alongside full-time VCs. Teams are rated on AI nativeness and market savviness, and the weaker side is de-risked with advisors. He expects a shift from AI-native teams winning obvious "consensus" applications after ChatGPT to market-savvy teams with non-consensus insight. He argues builders who also read papers beat a dedicated research PhD on the team.
• Portfolios need both heat-seeking and truffle-hunting bets. Johnson says hot deals the market already loves cost more, and a vintage's hot trends are rarely its best returners. Cheaper non-consensus bets are where big wins come from. He wouldn't steer founders away from crowded "red oceans", citing Lovable reportedly disrupting Cursor and Windsurf. AIX's second fund skipped climate because it found too few early-stage AI-climate deals and "atoms are more expensive than bits". The fund targets enterprise applications and tech-bio instead.
• Behaviour change, not technology, is enterprise AI's bottleneck. Johnson says WorkHelix, an AIX portfolio company he discloses, maps an enterprise's tasks from raw data to generative-AI solutions and proves gains in A/B tests. He reads investor rollups of service businesses, and professional-services line items at AI startups, as responses to that friction; one CEO told him they had to mandate AI every Friday. His fastest route to value is to fill open job requisitions rather than make current staff more productive, since teams resist tools that lead to a reduction in force.
• Enterprise change will be gradual and uneven. The host forecasts superhuman ability on narrow tasks within three to five years, but enterprises largely unchanged because of human friction. Johnson expects a spread: one 5,000-person company told him it would be 3,000 in two to three years, as AI takes over the 30–40% of engineering doing patches. The host reports, at the time of recording, about 6,000 Microsoft layoffs, mostly software engineers. Johnson's career advice is to advocate for AI now, and longer term to move toward human-to-human roles, since humanoid robots are two or three research breakthroughs away.
• Scarce frontier talent drives acqui-hires, which also protect investors. Johnson puts frontier AI practitioners at "single-digit thousands", so compensation is skyrocketing. He sees talent acquisitions as the opportunity cost founders must outrun, and as downside protection for investors; one AIX team went from operating to acquired by Google in about three days. Routes into that group, he says, are moving from engineering into labs like FAIR or DeepMind, or a PhD at schools such as CMU, Berkeley, Stanford, NYU, MIT or Toronto.

2025-06-09 · open the original ↗

• Overused practice datasets count against a portfolio. Asked by a listener what employers hold against newcomers, Smith names the Titanic, Iris and MNIST datasets. For analysts he adds Google's bike-share "capstone", each done by "probably literally millions of people". A project a few hundred people have done is fine. MNIST digit classification alone won't land a computer-vision job.
• A bare GitHub profile is not a portfolio. Smith admits the claim is deliberately provocative. He argues GitHub is a code repository, messy by default, and that recruiters give a portfolio 10 seconds before granting 60. A sparse or fork-filled profile wastes those seconds. He now thinks even his GitHub Pages site was overkill and suggests a simple builder such as Squarespace or Carrd.

2025-06-08 · open the original ↗

• Choose projects you care about, or they won't get finished. Smith says side projects die for lack of motivation, so they should tie to a hobby or a target employer. One killer project aimed at a company like Meta could be enough on its own.
• One SQL project and one data-viz project are enough to start. Smith calls waiting for five projects before applying a trap: "progress over perfection". More projects only help if they don't eat too much time.
• Personal-data projects motivate but can be costly to collect. His 2021 bootcamp built lessons around personal portfolio projects, such as a Looker Studio dashboard of phone screen time and Python on the Spotify API. He notes that Apple won't export screen time, while downloading a CSV into Tableau is far faster.
• An analytical mindset in a non-data job makes résumé material. A former student who was a delivery driver A/B-tested the texts she sent to get into gated buildings. Smith says that became a résumé line ("A/B testing marketing messages") that made her LinkedIn read like an analyst's.

2025-06-07 · open the original ↗

• A project makes a résumé keyword tangible and flips the interview. The host argues a working demo tied to a real passion matters far more than claiming PyTorch, and says he doesn't see how such a candidate could miss an entry-level job. Smith adds that a project supplied before or during an interview replaces "mystery questions" with questions about your own work.
• Public project work can bypass the interview entirely. Smith says he applied three or four times for a Utah Jazz internship without success. With about 60,000 LinkedIn followers, he then posted a project scraping and visualising the team's season shot data and tagged its analytics staff. He says he got the internship with no interview.

2025-06-06 · open the original ↗

• Agents digitise skills, where RAG chatbots unlock proprietary data. In a clip from episode 887, Dell's global CTO and chief AI officer John Roese names an enterprise's two sources of differentiation: proprietary data and unique skills. He says almost all of Dell's production generative-AI projects tackle the first, via RAG. Agents tackle the second by taking an objective and acting autonomously, with the human "on the loop" rather than in it. He frames them as narrow, geofenced skills like driverless airport trains, far from AGI.
• Polars proved itself through a small benchmarked rewrite. In episode 885, the co-authors of the O'Reilly book Python Polars: The Definitive Guide describe moving a pipeline off pandas and R at Dutch grid operator Alliander. The pipeline ran on a single AWS instance with over 700 GB of RAM. A benchmarked rewrite of one small piece won over a sceptical team, and one job shrank from about 500 GB to 40 GB. They say production use exposed gaps they fed back as issues, such as weighted random sampling.
• VR training can scale scarce teachers and risky practice. In episode 889, SEEK founder Mary Spio says one instructor can teach at scale in VR, against a global teacher shortage she cites as 4.6 million. She says an electric-aircraft client needs about 100,000 pilots soon, so a million must be trained. She adds that the brain forms memories in VR as if the experience were real. Her CPR training for new mothers and Baptist Health's nursing residency aim to build the confidence nurses lack, since the average nurse is 50.
• Conversational analytics depends on harmonised, monitored data. In episode 891, Adverity's Martin Brunthaler says data must be complete and aligned across sources, with consistent types and UTC timestamps. It also needs quality monitors, anomaly detection, a retained raw layer and a data dictionary with lineage. His pipeline classifies the question, prompts a model with context, generates and validates SQL, runs it and summarises. A benchmark team tests responses against expected answers, and he says models are converging, so differentiation lies in how components are composed.

2025-06-05 · open the original ↗

• Learn data tools in order of demand and ease. Smith's "Every Turtle Swims Past" ladder (Excel, Tableau, SQL, Python) ranks tools on how in demand and how easy they are. Excel is familiar and Tableau is drag-and-drop. SQL is "probably the most in demand" and easier than Python, which comes last for its steep learning curve.
• Getting hired fast beats learning everything first. Smith's philosophy is to get a foot in the door as soon as possible. The field changes constantly, and on-the-job learning is the most relevant and is paid.
• AI speeds coding but doesn't remove the need to program. Smith says AI gets him 60–70% of the way on code, but without programming knowledge you can't see where it is wrong. He sees it as a tool for professionals, not a replacement. The host uses Claude for most daily work and pays $200 a month for OpenAI's deep research, which he says saves him thousands of dollars of time.

2025-06-04 · open the original ↗

• A dangerous, low-paid lab job drove him into data. As a chemical lab technician, Smith got hydrofluoric acid on an ungloved hand and washed it off in time. He decided he wouldn't risk his life for near-minimum wage, and wanted the job of the company's data scientist instead.
• An internal pivot is the best route for career changers. Smith moved into a junior analyst role at the same employer, at 25%, then 50%, then 100% of his time. He argues a known, trusted employee is less of a risk than an outside hire without data experience. He credits the firm's size, under 15 people, and doubts it would have worked at ExxonMobil, where he later worked.

2025-06-03 · open the original ↗

• Career changers reject themselves before employers do. Smith says "the number one hiring manager that rejects you from jobs is yourself". Data is everywhere, so past roles can be framed as data experience; at around 50% of a job's requirements, he says, "just apply". The host adds research he has read that women are less likely than men to apply without meeting every criterion.
• An internal pivot is the lowest-risk route into data. After a near-miss with hydrofluoric acid as a lab technician, Smith moved into a junior analyst role at the same company, at 25%, then 50%, then 100% of his time. He says employers who already know and trust you see less risk. He credits the firm's size, under 15 people, and doubts it would have worked at ExxonMobil, where he later worked.
• Learn data tools in order of demand and ease. His "Every Turtle Swims Past" ladder (Excel, Tableau, SQL, Python) starts with what is in demand and easy. SQL is "probably the most in demand", and Python comes last for its steep learning curve. The aim is to get hired fast, since on-the-job learning is best. He says AI gets him 60–70% of the way on code, but you must know programming to see where it is wrong.
• Portfolio and network weigh as much as skills. Smith rates skills, portfolio and network as equal thirds, and has considered reversing the order. His line is "resumes talk, projects walk". After three or four rejections, he won a Utah Jazz internship with no interview by tagging the analytics staff on LinkedIn in a project scraping their shot data. For analysts, one SQL project and one data-viz project on a real interest is a start. He would avoid overused datasets (Titanic, Iris, MNIST, Google's bike-share capstone), and argues a bare GitHub is no portfolio when recruiters give it 10 seconds.
• Run the job hunt like an experiment. At about 14 hours a week Smith says a learner can be fully skill-ready in roughly 12 weeks, applying from week four to six; many of his students are hired in under 100 days. His analysis found about 15% of data jobs fully remote, against far more who want one. He has students A/B test résumés and keep a job tracker; one tracker showed 75% of applications going through LinkedIn Jobs, so he told her to switch platforms.

2025-06-02 · open the original ↗

• Build instead of overthinking architecture. Brunthaler, a serial entrepreneur and Adverity's CTO, says his teams have fallen victim to overthinking. AI coding tools now make it quick to get something running, though he says he is not endorsing vibe coding.
• Use proven "boring" tech and keep talking to customers. He would compose an exciting use case from technology that works. He calls customer feedback the most critical input, and says Adverity has been customer-centric from the start.

2025-06-01 · open the original ↗

• Data work will shift from mechanics to strategy. Brunthaler expects many workflows that still need manual steps to be fully automated. People would then focus on the business outcome and the "why" of data rather than the "how". He hedges that five years is a very long horizon in a field where he thinks in half-years.
• Separating compute from storage will reshape the data stack. He describes raw data held in standardised object storage, with the industry settling on Iceberg as the format. Different query engines, and likely generative AI, would connect to it depending on purpose, letting businesses pick the right engine for each case.

2025-05-31 · open the original ↗

• "Agentic AI" has been stretched to cover things it isn't. Brunthaler says the term, like "AI" before it, is applied to statistical and machine-learning features with no generative model in them; his working definition is a ChatGPT-style model given access to functions so it can act, or retrieve information, on the user's behalf. The host revises his own broader usage and proposes "an LLM that can take actions", which Brunthaler accepts, adding that a definition that is too broad sets expectations systems cannot meet.
• Users don't yet trust AI to decide fully on their behalf. Brunthaler holds that some level of control and monitoring has to be built in, even where an agent can call an API, use the result or trigger an action without user interaction.
• The industry is now in a standards-setting phase for agents. Brunthaler points to the recent hype around MCP and expects more around A2A, and says Adverity, like many SaaS products, will offer MCP so a company's agent of choice (Gemini, OpenAI or another) can hook into it. He describes the integration as early, currently needing Claude Desktop, with use cases such as listing a customer's connected data sources, checking they are properly connected and creating re-authorisation links.

2025-05-30 · open the original ↗

• Enterprise generative AI has entered the "trough of disillusionment". Using Gartner's hype-cycle term, the host cites S&P Global survey results that 42% of companies are abandoning most of their generative AI projects, up from 17% a year earlier, and reports that Klarna admitted it went too far replacing customer-service staff with AI and is rehiring humans; Gartner analysts, he says, expect the trough to last until the end of the year.
• Consumer adoption is booming even as business adoption stalls. The host relays Sam Altman's figure of 800 million weekly ChatGPT users, double the February level, against corporate obstacles he lists: data trapped in silos and legacy IT, a shortage of technical talent, and brand and privacy risk from bots that make damaging mistakes.
• Hyperscaler AI spending is outrunning visible business returns. He cites Pierre Ferragu of New Street Research that the combined capex of Alphabet, Amazon, Microsoft and Meta is on track for 28% of revenues this year, against 12% a decade ago. He notes that Nadella and Pichai showcased token counts rather than sales or profit from AI, and that most AI cloud revenue comes from labs and startups the same giants fund, which he calls a circular economy.
• The hyperscalers' internal AI uses are where returns are showing. He lists Google's AI search summaries reaching over 1.5 billion people monthly, Meta's Llama-based ad tools, Microsoft's workplace apps and GitHub Copilot, and Amazon's recommendations and logistics, and links Microsoft's recent 6,000 layoffs, reportedly many software engineers, to AI coding tools. He presents Apple's postponed LLM rebuild of Siri as the cost of moving too slowly.
• The bottleneck is applying existing capability, not building smarter models. The host calls this a "capability overhang" and, citing Microsoft CTO Kevin Scott, says agents still need better memory and protocols such as MCP. His own conclusion is that the trough is an opportunity: the slope of enlightenment follows, and practitioners who keep solving integration, talent and risk problems will come out strongest.

2025-05-29 · open the original ↗

• Natural-language queries replace hunting for the right dashboard. Brunthaler says Adverity's Data Conversations lets a user ask a question instead of finding a dashboard or asking another team for data, and returns a table that can go straight into a Google Sheet or be materialised as a Snowflake or BigQuery table. He says the company's marketing subject-matter expertise is built in, so a question like "best performing campaigns" gets connected to the right data.
• Conversation allows exploration that BI drill-downs cannot. Brunthaler argues that branching into other data sets to understand why a trend is happening is possible in a conversation but not with an explore function in Looker or in Adverity's own earlier tooling.
• Enterprises need a tool they can manage, own and audit. Brunthaler presents governance as a core requirement: a single governed tool rather than a mix of connected SaaS tools that cannot meet current regulatory demands.
• Conversational data tools shift analysts' work rather than removing it. Brunthaler says they free analysts and data engineers from ad hoc data-pull requests to focus on strategic work, and that the role will move toward governing these tools and checking their answers are correct. He sees no jobs at risk, only changing profiles.

2025-05-28 · open the original ↗

• Conversational data access depends first on underlying data quality. Brunthaler says the data set must be complete and harmonised across sources, with aligned data types and formats (all timestamps in UTC, for example), alerts when something goes wrong, and problematic data kept out of production. Adverity's platform, he says, has marketing-specific monitors such as campaign naming-convention checks, plus anomaly detection.
• Keep a raw layer so transformations can be redone. Brunthaler describes a multi-layer approach in which the raw data set is always retained as the starting point for revising transformations, with AI assistance in composing them. He adds that data dictionaries, descriptions and lineage are critical for the conversational interface to answer correctly.
• Response quality is benchmarked continuously against expected answers. Brunthaler says a dedicated team tests against a predefined set of expected responses while features ship roughly weekly. Adverity is currently committed to one model but plans separate models for SQL compilation, pre-flight qualification of a question, and the conversation itself.
• Models are converging, so the advantage is in assembly. Brunthaler says there is no trade secret in the LLMs, whose leaderboards change every other month, and describes the pipeline as plain: classify the question, pick a model, supply a system prompt and model metadata, generate SQL, validate and run it, then analyse the result into an answer. The payoff he claims is a near-real-time table in place of a ticket that used to take two weeks.

2025-05-27 · open the original ↗

• Dashboards have failed at democratising data; conversation is the next step. Brunthaler, Adverity's co-founder and CTO, says dashboards are built once, rarely reviewed, mistrusted, and give no way to ask follow-up questions, so users end up back in the source systems. He says Adverity's Data Conversations, launching at recording, answers natural-language questions with an actionable table and allows branching into other data sets in a way a Looker-style explore function cannot.
• A conversational layer only works on a governed, high-quality data foundation. Brunthaler lists the prerequisites: complete, harmonised data with aligned types and formats (UTC timestamps), marketing-specific monitors such as campaign naming conventions, anomaly detection, and a retained raw layer so transformations can be redone. He says Adverity's data dictionary, linking each attribute to its source and meaning, is what grounds the model's answers.
• LLMs are converging, so the edge lies in assembly and testing. Brunthaler says there is no trade secret in the models, whose leaderboards turn over every other month; the pipeline classifies a question, picks a model, adds a system prompt and metadata, generates and validates SQL, runs it and analyses the result. A dedicated team benchmarks responses against a predefined expected set, and Adverity plans to move from one model to separate models per stage.
• Rebuilding a modular "modern data stack" is wasted engineering effort. Brunthaler argues Adverity replaces a typical BigQuery or Snowflake, dbt and Power BI/Looker/Tableau chain by delivering harmonised marketing data ready for BI, running dbt for customers and keeping up with frequent changes at sources such as Google and Facebook. He calls agentic AI an overused term, defines it with the host as "an LLM that can take actions", and says Adverity will offer MCP.
• Within five years, work shifts from data mechanics to strategy. Brunthaler forecasts that manual workflows will be automated and that separating compute from storage, with raw data in object storage in the Iceberg format the industry has settled on, will let businesses pick query engines, and eventually generative AI, by purpose. He hedges that five years is long when his own horizon is about half a year. His founder advice: don't overthink, use boring proven tech with AI coding tools (not vibe coding), and keep talking to customers.

2025-05-26 · open the original ↗

• Live out of your imagination, not your circumstances. Spio, summing up her book It's Not Rocket Science: Seven Gamechanging Traits for Uncommon Success, points to dreaming of walking on the moon while growing up in Ghana at a terrible time. Her advice is to start before conditions are perfect, since things fall into place once the journey begins.
• When investors judge the founder, not the business, change venue. Spio recounts Silicon Valley investors remarking that she didn't look like an engineer, one handing her a book on how engineers dress, and VCs saying they had no fund for people of colour or no diversity fund yet. Quoting the line that sometimes you need a change of perspective, not patience, she says her first big investor came in New York, from backers focused on her revenue. The host's research says she bootstrapped CEEK on credit cards through enterprise sales cycles of a year or more.
• Go where you are valued. Spio says a satellite employer (Intelsat, then PanAmSat) where her boss was unfriendly lost her content-distribution work when Boeing recruited her; she sold her patents to Boeing, the technology grossed it billions and was sold for a billion, and her former department was later laid off, with the old boss asking her for a position.

2025-05-25 · open the original ↗

• AI will tailor immersive experiences, even narratives, to each person. Spio, who the host says holds over 10 patents across AI, digital cinema, spatial audio and extended reality and invented the first 360-degree 4D audio headphones, says this customisation is already possible and extends to education, matching different learning styles. She calls current capabilities the tip of the iceberg.
• Personalised VR makes manipulation a real risk, the host argues. He puts it to Spio that bespoke, emotionally targeted content in VR, where the brain cannot tell it from reality, could be used to shift people's political preferences on top of today's deepfakes.
• Displaced workers will move into curating and validating AI content. Spio quotes Elon Musk that mass unemployment from automation is a certainty and a figure she attributes, uncertainly, to the IMF that 40–60% of today's jobs will be affected. She expects those workers to curate AI experiences and to verify what content is real, because plausible AI output means people can no longer tell fact from fiction, as with campaign videos.
• Spio proposes blockchain validators to certify content accuracy. She says CEEK is building a platform with blockchain master nodes and validators dedicated to checking content, so users know what they consume has been validated. She concedes it will not be 100% accurate but argues it beats doing nothing.

2025-05-24 · open the original ↗

• VR headsets designed around men's eyes made women sick. Spio says that at demos, including one at Universal Music, most of the women who tried a headset ran to the bathroom to vomit. She traced it to interpupillary distance: headsets were built with a lower limit set at men's range, which sits at or beyond women's upper limit, pulling the eyes apart and causing strain and headaches.
• Men and women judge depth from different visual cues. Spio says men rely more on an object's size to judge distance, while women rely on detail, so content rendering based on motion parallax and shape-from-shading felt unnatural to women. She says she designed a headset that resolves these issues, and her platform uses predictive rendering to keep content comfortable for everyone.

2025-05-23 · open the original ↗

• Small models now match what once took enormous ones. The host relays the Stanford HAI 2025 AI Index: in 2022 Google's PaLM, at 540 billion parameters, was the smallest model above 60% on MMLU, and by 2024 Microsoft's Phi-3-mini cleared the same bar with 3.8 billion, a 142-fold reduction in two years.
• Inference for a given capability has become dramatically cheaper. Per the report, GPT-3.5-level performance (65% on MMLU) fell from about $20 per million tokens in November 2022 to 7 cents by October 2024, using a Gemini 1.5 model, a 280-fold drop in roughly 18 months; depending on task, prices fell between 9 and 900 times a year from 2023 to 2024.
• Agents beat experts on short tasks but lose on long ones. On the RE-Bench benchmark launched in 2024, top AI systems scored four times human experts at time horizons of 2 hours or less, while humans outscored AI 2:1 at 32 hours. The host predicts that within a year agents will outperform humans on tasks of a dozen hours or more.
• Corporate AI adoption jumped sharply in one year. Survey data in the report show company-wide AI use rising from 55% in 2023 to 78% in 2024, and generative AI use in business functions from 38% to 71%.
• US private AI investment hit a record while China's kept falling. The report puts US private AI investment at $109 billion in 2024, about 20 billion above the 2021 peak, and Europe's at $19 billion. China's has fallen every year since peaking at around $25 billion in 2021, to $9 billion.

2025-05-22 · open the original ↗

• VR lets one instructor teach at a scale classrooms cannot. The host cites a US shortage of 400,000 K–12 teachers and ageing university faculty; Spio cites a stat she "saw somewhere" of a 4.6 million shortfall across Europe and North America, with Africa needing "a miracle". She says CEEK lets a single person present a course virtually to vast audiences, and the host notes she has described one professor teaching 100,000 students.
• Electric-aircraft pilot training is a concrete demand case. Spio says putting 100,000 eVTOL pilots in service within a few years means training a million, which cannot be done in expensive, risky real aircraft. She says CEEK is working with a leading electric aircraft company with a massive backlog, and puts the operating cost at about 300 an hour against about 4,000 an hour in fuel for conventional helicopters.
• The brain stores VR practice as real experience. Spio argues the brain cannot yet tell VR from reality, so trainees build memories and show up on day one with hours of practice already logged. She cites CPR training built for a children's hospital, where new mothers felt more confident than after watching a video.
• Nurse turnover is a confidence problem VR can address. Spio says Baptist Health pursued a VR nursing residency because the average nurse is 50 and many new nurses leave within a year or two out of fear of hurting someone, not incompetence. She adds that some hospitals train intubation on paid low-income people, the homeless or elderly people with Alzheimer's, at around $50, which VR would replace.

2025-05-21 · open the original ↗

• CEEK stores users' content as blockchain assets they own. Spio, CEEK's CEO and CTO, describes it as a social platform where every post or video is a digital asset on the blockchain backed by the platform's own token, tradable and collectable, viewable with or without a headset (Apple Vision Pro, Oculus), with users keeping rights to their data. The host says it is backed by Universal Music, Meta, Apple and Microsoft and hosts experiences from Lady Gaga, Bon Jovi and Dwyane Wade.
• AI music is the marketplace's best seller, and rights are what's sold. Spio says AI music is probably the most purchased item, bought by film supervisors and labels, because the marketplace sells streaming, sale or licensing rights rather than downloads. She argues this works because AI music usually has a solo creator owning the full rights, making smart-contract sales simple, unlike traditional music's many parties. She thinks the best AI music still has humans finessing it.
• AI vocal enhancement trains the singer's natural voice. Spio says she uses AI to produce beats and enhance her vocals, "autotuning on a whole different level", and that hearing her own voice corrected has improved her natural singing fast enough that she now releases music under her own vocals rather than writing for others. She describes AI as letting her mix genres such as EDM and Afrobeat.

2025-05-20 · open the original ↗

• CEEK is a paid, professional alternative to ad-based platforms. Spio, CEEK's CEO and CTO, describes a social platform storing content as blockchain assets users own, viewable with or without a VR headset. She argues ad-based platforms serve advertisers and work for maybe 3–5% of creators, and contrasts a million YouTube views of eVTOL training earning perhaps $3,000 with a thousand people paying 5,000 each. She cites BTS generating 90 million from a paid concert in three days, not on CEEK, and says a flood of content against finite ad inventory is squeezing ad-based platforms.
• VR training scales scarce skills and builds real memories. Spio says putting 100,000 eVTOL pilots in service within a few years means training a million, and CEEK is working with a leading electric aircraft company; she puts its cost at about 300 an hour against about 4,000 in fuel for helicopters. Because the brain cannot yet tell VR from reality, she argues, trainees arrive with experience already built, as in CPR training for new mothers and a Baptist Health nursing residency aimed at turnover she attributes to low confidence, not competence.
• Headsets designed around men's eyes made women nauseous. Spio traced women vomiting at demos, including at Universal Music, to interpupillary distance ranges set at men's lower limit, at or beyond women's upper limit, and to men judging depth more by size and women by detail. She says she built a headset that resolves this and uses predictive rendering on the platform.
• Personalised immersive content needs verification, Spio argues, via blockchain. The host raises VR deepfakes tuned to shift political views; Spio cites Elon Musk on certain mass unemployment and an IMF figure, as she recalls it, of 40–60% of jobs affected, and expects displaced workers to curate and validate content. She says CEEK is building blockchain master nodes and validators to check content accuracy, conceding it won't be 100%.
• Win on value to the ecosystem, not on funding. Spio says CEEK beat better-funded rivals at Universal Music by understanding entitlements and letting the label drag and drop its log files to port a multi-million library, distributed to Oculus, Apple Vision Pro and HTC Vive without separate builds; Meta funded it after it hosted thousands of people in one VR environment. Her success philosophy: live out of your imagination rather than your circumstances, start before it's perfect, and go where you are valued.

2025-05-19 · open the original ↗

• Viable quantum computing will be a bigger disruption than ChatGPT. Roese argues that because qubits let a machine evaluate many permutations at once, whatever AI is state of the art on that day will become "three, four, five orders of magnitude faster and better", in training and in agents' inference and reasoning. He notes quantum computers only do certain kinds of math well, and that machines able to break asymmetric cryptography do not exist yet but are coming.
• AI and quantum computing are accelerating each other. Roese says AI is now being used to build, interface with and program quantum computers, speeding a historically slow path, while viable quantum machines will in turn give AI a far more capable computing infrastructure.
• Enterprises should prepare now, though the date is a few years out. Roese puts quantum utility, then supremacy, "probably not for a few more years" but "not decades from now". A lay person need not worry yet, he says, but enterprises should act as if they could replay the November ChatGPT launch knowing in advance it was coming.

2025-05-18 · open the original ↗

• Leaders owe workers concrete answers about which jobs AI creates. Roese says displacement fear is the main source of angst about AI, and saying "technology always creates jobs" is not enough for someone who expects to lose theirs. He says Dell drew up about a dozen specific future roles, published a few in an end-of-2024 blog, and is seeing them form organically inside the company.
• "Software composers" will decide what AI should build. Roese argues AI cannot make architectural judgments by itself, such as microservices versus monolith, cloud native or not, or whether to follow the twelve-factor approach, and cannot act without a clear specification. He says he is asking universities for graduates who understand software architecture, can use coding assistants and need not write the code themselves.
• "Thermal plumbers" combine disciplines GPU clusters now need. Running direct-liquid-cooled GPU clusters, Roese says, requires someone who understands computer hardware as well as thermodynamics and fluid dynamics to manage the thermal envelope. He says these are few but essential, and the role is already forming at large clusters.
• "AI explainers" will deliver machine insights to people empathetically. Roese's examples are a genomics result showing risk of Alzheimer's or Parkinson's, an automated performance review, and a student the data predicts will fail. Each needs a human who understands the data and how the AI reached its conclusion. He calls this a large, non-technical path, possibly "the call center of the future".
• Construction is the biggest job creator of the AI cycle. As a caveat to all of the above, Roese says the AI infrastructure build-out, employing construction workers, plumbers and electricians, is bigger than the US infrastructure bill of several years ago, a private-sector public-works project that will last a long time. He still warns the disruption is moving fast.

2025-05-17 · open the original ↗

• An agent's value lies in the collective, as with people. Roese says Dell built its first autonomous agents over a year before recording, a two-agent system writing research reports, and every system since has had at least two agents and a human. One system flexed to 1,600 agents on a problem because agents could "hire" others when they lacked a skill.
• The industry has no framework for agents working together. Roese says the message format is settled (JSON, human-readable text), but how to authenticate and authorise an agent, share knowledge across companies, or prompt a Dell agent and a partner's agent jointly is unresolved. He calls Anthropic's MCP a well-thought-out answer for models talking to data, and Google's Agent2Agent, announced the day before recording, promising for authentication and interworking.
• De facto standards will arrive within the year, by consensus. Roese says he is "100% confident" that before the end of the year there will be de facto approaches to trustworthy agent interaction, emerging from industry activity rather than a standards body or a pure open-source project. He expects them to stay ring-fenced at "level four autonomy", not handling every corner case.
• Existing security tools can be reused for agent ensembles. Roese argues authentication, authorisation and knowledge sharing need choosing among existing tools, not inventing, citing knowledge graphs, confidential computing and (partially) homomorphic encryption, which processes data without seeing it.

2025-05-16 · open the original ↗

• Innovation culture comes from not asking permission. Pell, director of the Microsoft Garage, says that since Satya Nadella became CEO about 10 years ago Microsoft has tried to build curiosity and experimentation into everyday work rather than offering 20% time. In practice that means telling people not to ask permission, just try it and bring back data proving whether it works.
• AI is an exoskeleton; agents package it as a trusted co-worker. Pell describes AI as an Iron Man suit giving capabilities people would never have time to master. He predicts assistants will shift quickly from passive to proactive, interjecting during work ("did you really mean this part?"), and says trust must be earned through repeated correct behaviour without users falling "asleep at the wheel".
• Early imperfection nearly killed GitHub Copilot. Pell says engineers dismissed the first versions over syntax errors, and that the tasks Garage clients are shocked by, such as market research, competitor analysis and go-to-market strategy, can drop from three days to 30 seconds.
• Teams of agents on ambiguous tasks are not close yet. Pell says chaining tasks is the current state of the art, and coordinated "swarming" of agents is not near. The host notes frameworks such as CrewAI and argues a 1% per-agent error rate compounds across 10 agents; Pell answers that review agents and orchestration could instead lower errors.
• Soft skills become the job as hard skills automate. Pell says graduates arrive smart but unable to communicate, debate or think critically in the moment, as with the Italian-style oral exams many American students would fail, and argues AI should be used to become a better teammate and collaborator. He holds AI is already ubiquitous, in Netflix and Amazon recommendations, and that within a year nobody will talk about it as AI. His advice to non-coders: take no class, talk to a chatbot like a person and start with small wins.

2025-05-15 · open the original ↗

• An enterprise is differentiated by its proprietary data and its unique skills. Roese argues these two are what AI must be applied to. Generative AI went after the first: RAG chatbots that embed proprietary data, such as service information, in a vector database and make it answerable, which he says almost all of Dell's production projects do, at industry scale for about a year.
• Agents digitise narrow skills, not whole humans. Roese says agents go after the second source, taking an objective and reasoning through it autonomously to deliver an outcome equal to or better than a human's on that skill. Like geofenced self-driving cars or driverless airport trains, today's agents do one job; a well-rounded entity with ethics and morality is AGI/ASI and "a long ways away".
• Agentic AI moves the human from in the loop to on the loop. In Roese's framing, reactive generative AI is a tool with the human doing the work, while with agents the human sets objectives and delegates, and work may stay "below the machine line" indefinitely. He says the architecture differs too: the LLM is the brain, with a knowledge graph holding memory and learned skills, and tool use and function calling giving it a body to act and perceive.

2025-05-14 · open the original ↗

• ROI is the first question before funding any AI project. Roese, Dell's global CTO and chief AI officer, says AI is useful only in service of a measurable outcome tied to an improvable process: profit, revenue, margin and cost for a company, educational outcomes for a university. AI that gives salespeople more time with customers, which Dell did, counts; AI that makes them slightly happier is meaningless unless measured.
• Ask where AI should be applied, not where it could be. After the ChatGPT launch Dell received about 800 generative AI ideas, many of them, Roese jokes, inspired by Star Trek, and found it could not pick winners from them without context. Dell flipped the process: start from what it measures (profit, revenue, cost, regulatory risk), pick four areas (supply chain, sales, services and engineering), and find the biggest lever in each, such as sellers' content preparation and core coding in engineering. The host's research says the 800 were narrowed to eight.
• Every enterprise is stuck at the same starting problem. Roese says every customer he talks to has hundreds of ideas and struggles to choose where to begin. Running 800 concurrent AI projects is impossible even for the largest companies, and debating them all leaves nothing in production.

2025-05-13 · open the original ↗

• Fund only AI that moves a measurable business outcome. Roese, Dell's global CTO and chief AI officer, says Dell's roughly 800 post-ChatGPT ideas could not be ranked without context, so it asked where AI should apply: profit, revenue, cost and regulatory risk, across four pillars (supply chain, sales, services, engineering). His gate to production is a material ROI, alignment with how Dell wants to run in future rather than "a big blanket" over bad processes, then technical viability and security and compliance. That gate is how he says companies escape "PoC prison".
• High-ROI first projects start a flywheel that funds the rest. Roese argues later projects reuse the first one's technology and so cost less and ship faster, while starting with a goodwill project produces no fuel; Dell left G&A functions for later. Sales went last because its data needed fixing first, after which Dell Sales Chat went live for 20,000 sellers and, he says, exceeded expectations. Before ChatGPT, Dell ran 500–1,000 AI experiments a year, 80% with zero business impact, which he defends as building familiarity.
• Agents digitise narrow skills, putting humans "on the loop". Roese says enterprises differ by proprietary data, which RAG chatbots have unlocked and which accounts for almost all of Dell's production projects, and by unique skills, which agents digitise. Like geofenced self-driving cars, agents do one job autonomously from an objective; whole-human AGI/ASI is "a long ways away". He says agent architecture differs too: an LLM brain plus a knowledge graph for memory and tool use to act.
• Agent interworking standards will emerge by consensus this year. Dell has run multi-agent systems since its first two-agent research-report writer, once flexing to 1,600 agents that could "hire" others. Roese says authentication, authorisation and cross-company knowledge sharing are unsolved, praises MCP for model-to-data and Google's Agent2Agent, announced the day before recording, and is "100% confident" de facto approaches arrive before year-end, ring-fenced at "level four autonomy" and reusing tools such as homomorphic encryption.
• AI will create named new jobs, but construction is the biggest. Roese lists software composers who decide architecture without writing code, "thermal plumbers" who manage liquid-cooled GPU clusters, and AI explainers who deliver AI findings empathetically, as with genomic risk results, and says such roles are forming inside Dell. Above all, he says, the AI infrastructure build-out exceeds the US infrastructure bill of several years ago. He also forecasts that viable quantum computing, "a few more years" away but not decades, will be a bigger disruption than ChatGPT, making AI three to five orders of magnitude faster.

2025-05-12 · open the original ↗

• The authors benchmarked Polars on GPUs themselves, with vendor hardware. Janssens says Nvidia approached them on LinkedIn and Dell supplied a machine in its US lab where five professional workstation cards, including the RTX 6000 Ada generation, could be swapped. The pair insisted on producing their own numbers rather than copying promotional material, and the results are in the appendix of their book. The host notes Dell and Nvidia are two of the podcast's biggest 2025 sponsors.
• Polars' GPU engine sits after its query optimiser. Nieuwdorp explains that Nvidia's RAPIDS team, maker of the GPU dataframe library cuDF, did not simply map the Polars API onto cuDF functions, because much of Polars' speed comes from its optimiser. Instead it built, with Polars, a GPU engine that takes the optimised plan; it was in open beta at recording.
• GPU gains start on modest data, even on smaller cards. Nieuwdorp says data transfer to the GPU adds overhead, so speed-ups appear as data grows, but already from around 1 GB, sooner than the promotional material suggested. Even cards with fewer processors showed big speed-ups.

2025-05-11 · open the original ↗

• uv replaced Poetry and Docker for the book's reader environment. Nieuwdorp says the authors started with Poetry plus separate Python-version tools and considered Docker for the companion notebooks repo, but settled on uv because readers need only install it and run uv sync, which sets up the right Python version and dependencies on any system.
• Tooling performance is a feature in itself. Nieuwdorp places uv in the trend of Rust-based tools, like Polars, and says it can be more than 10 times faster.
• uv's speed makes throwaway environments practical. Nieuwdorp says uv can create an ephemeral environment for a single command, which he used to benchmark successive Polars versions. He found a roughly 10% slowdown around version 1.2-something, traced it to two queries, and used a git-bisect script to identify the commit. The Polars team refactored that code the same week, which resolved it, though they had not pinned down the cause.

2025-05-10 · open the original ↗

• Tables are underserved by Python's visualisation tools. Janssens says that despite a myriad of plotting packages, Great Tables, created by his colleagues Rich Iannone and Michael Chow, is the only one he knows for producing presentation tables with colour, currency formats and mini graphs. Polars exposes it through the df.style accessor.
• Format how data is shown; never change the data. Janssens objects to a post he saw that added dollar signs by converting numbers to strings. Styling should be a separate representation layer, he argues, so the underlying floats and integers stay intact for later calculations. He corrects the host's reading that the point is computational cost.

2025-05-09 · open the original ↗

• Nvidia packages models as microservices so they can be swapped. In episode 883, Nvidia's Sama Bali explains that NIM (Nvidia inference microservices) containerise open, proprietary and Nvidia models pre-tuned to the GPU, because models turn over every couple of months, such as Llama 3 to 3.1 to 3.2. Dell's Logan Lawler says a NIM loads with one line of code. Bali says CUDA libraries such as RAPIDS cuDF can speed pandas- or Polars-style preprocessing by 100x on RTX 6000 GPUs without code changes, and cuML up to 50x for scikit-learn tasks.
• AWS's custom chips rest on Annapurna Labs' Nitro work. Emily Webber of AWS (episode 881) says Amazon acquired Annapurna in 2015 to break up the monolithic hypervisor. Its Nitro system physically separates customer data from instance control, and every modern EC2 instance runs on it. She says more than half of new AWS compute is Graviton Arm CPUs, and calls Trainium 2 the most powerful EC2 instance for AI/ML, chosen mainly for price-performance and energy efficiency.
• Data scientists outnumber the engineers who deploy their models. Greg Michaelson, Zerve's co-founder and chief product officer (episode 879), calls the hand-off a problematic dependency where mistakes creep in. He says Zerve gives every canvas a Docker container so environments are shared, serialises every variable so a trained model can be pointed at for an API route, and deploys on serverless Lambdas, letting data scientists ship their own software.
• Heterogeneous integration scales chip performance beyond shrinking transistors. Kai Beckmann (episode 875) describes combining dies into one system, such as stacking memory or placing it beside a GPU (CoWoS-style structures) to shorten data transfer. He says it needs front-end-like precision, heat and power management, and new materials, which his company develops with AI-driven "materials intelligence" to cut unnecessary experiments.
• "AIPC": accelerated, individualised, private, cost. Shirish Gupta (episode 877) gives his mnemonic for on-device AI. His examples: an ER physician's report drafts that learned the doctor's style, data that never leaves the device, and a financial-services customer spending 15% of data-centre compute on developers' code generation, which it wants offloaded to laptop NPUs. He concedes cloud inference costs are falling fast but argues for "the right engine for the right workload".

2025-05-08 · open the original ↗

• Polars rescued a pandas pipeline that could not scale. Janssens says Alliander, the Dutch grid operator which runs electricity and gas infrastructure for a third to half of the Netherlands, needed to cut cost and process far more data, but its Python and pandas code was inefficient and ran on a single AWS instance with over 700 GB of RAM. He believes it was one of the first companies with Polars in production, over a year before Polars 1.0.
• Win over a sceptical team with one small benchmark. Janssens says the team was understandably wary of a library promoted by three consultants from Xomnia, where Polars was being developed. The consultants rewrote a piece of low-hanging-fruit pandas code in Polars and showed the speed and memory numbers. The full migration then went piece by piece, idiomatically rather than line by line, until R and pandas could be dropped as dependencies.
• Memory per job fell more than tenfold. Nieuwdorp puts it at about 500 GB down to 40 GB, and attributes the drop partly to Polars and partly to restructuring code along the way.
• Production use and book-writing exposed Polars' gaps. Janssens says real work surfaced missing features, such as weighted random sampling that pandas has, while filling in the book's reference tables revealed naming inconsistencies and missing operators. The authors fed these back to the project as issues and some pull requests.

2025-05-07 · open the original ↗

• Polars code reads as declarative paragraphs, not nested brackets. The authors say pandas code is full of brackets and often hard to reason about, whereas Polars expressions (in their book's metaphor, recipes whose operations are steps and whose functions are cooks) compose into queries that read like a paragraph. Users declare the end result and leave processing and optimisation to the engine.
• Polars was shaped by pandas' failures and strengths alike. Relaying conversations with creator Ritchie Vink, the authors say a formative frustration was pipelines crashing 20 minutes in on errors that could not be seen up front. Polars kept what Vink liked in pandas and borrowed from Spark, whose syntax it resembles, and from Rust. They stress they are not bashing pandas: without Wes McKinney's library there would be no Polars.

2025-05-06 · open the original ↗

• Polars' declarative style makes pipelines easier to read than pandas. Janssens and Nieuwdorp, authors of O'Reilly's Python Polars: The Definitive Guide (released on the recording date, 1 April 2025), say pandas code is bracket-heavy and hard to reason about, while Polars expressions compose into a query that reads like a paragraph, leaving optimisation to the engine. They credit pandas as Polars' foundation, and relay that creator Ritchie Vink was shaped by pipelines crashing 20 minutes in, borrowing from Spark and Rust.
• A production migration cut memory needs by an order of magnitude. At Alliander, memory per job fell from about 500 GB to 40 GB. The Dutch grid operator's pandas pipeline ran on a single AWS instance with over 700 GB of RAM. The authors won over a sceptical team by benchmarking one small rewrite, then migrated idiomatically until R and pandas were dropped. Nieuwdorp says the same time frame that once fit 25 simulation samples later fit the 50 stakeholders wanted, with code restructuring contributing alongside Polars.
• The GPU engine pays off on modest data sizes. With five Nvidia workstation cards in a Dell lab, both companies being podcast sponsors, the authors ran their own benchmarks for the book's appendix rather than reuse promotional figures. Nieuwdorp says Nvidia's RAPIDS team built an engine that takes Polars' optimised plan instead of mapping the API onto cuDF. Speed-ups appeared from around 1 GB, sooner than promotional material implied, and even on smaller cards.
• uv is fast enough to change how tools get used. Nieuwdorp says uv replaced Poetry and a planned Docker setup for the book's repo, since readers need only uv sync. Its speed allowed throwaway environments per command, which he used to benchmark Polars versions, find a roughly 10% regression around version 1.2-something, and git-bisect it to a commit.
• Format a table's presentation, never its data. Janssens says Great Tables, by his colleagues Rich Iannone and Michael Chow and exposed in Polars via df.style, is the only Python table-styling package he knows. Turning numbers into strings to add dollar signs breaks later calculations. He also recounts rewriting most of the visualisation chapter (16) when Marco Gorelli switched Polars' plotting backend from hvPlot to Altair, a feature then marked unstable.

2025-05-05 · open the original ↗

• AI development will reach many more non-developers. Dell's Logan Lawler says that, not being a developer when hired about a year earlier, he taught himself to pull SDKs from GitHub and train his own LoRAs for an animation studio. He expects that kind of access to become far easier and more common.
• AI will simplify small everyday tasks. Lawler's example is a RAG system over his 70-year-old mother's thousands of unsorted recipe cards that would return her pumpkin pie recipe on request. He notes someone would still have to set it up.
• Agents will coordinate with one another, then move into the physical world. Nvidia's Sama Bali, echoing Jensen Huang's GTC keynote (about five weeks before release, the host notes), describes factory agents passing word of a raw-material shortfall to the floor agent and then the transport agent so fewer trucks are booked. She points to Bay Area driverless cars visibly improving week by week as physical AI, and adds her personal hope for guardrails and regulation.

2025-05-04 · open the original ↗

• NIM microservices exist because models change every few months. Bali says Nvidia packages open, proprietary and its own models as containers with TensorRT-LLM-style inference optimisation, so developers can swap Llama 3 for 3.1 or 3.2 without rebuilding the pipeline; NIM stands for Nvidia inference microservice. Lawler adds that a NIM loads with one line of code and arrives already tuned to the GPU it runs on, so the only check is that the model fits in GPU memory, 96 GB on the Dell Pro Max machine discussed.
• Nvidia AI Enterprise layers training, guardrails and recipes on top. Bali lists NeMo for building, fine-tuning and guardrailing models, and AI Blueprints as reference workflows to which companies add their own data, which she calls their differentiation. She says NIMs can be prototyped free at build.nvidia.com and downloaded free through Nvidia's developer program.
• CUDA libraries accelerate data science without code changes. Bali says CUDA's parallelism can cut training from weeks to days. By her account, RAPIDS cuDF mimics pandas and Polars APIs to speed preprocessing by 100x on RTX 6000 GPUs with one added line, and cuML, announced at GTC, speeds scikit-learn tasks by up to 50x.

2025-05-03 · open the original ↗

• Data-science acceleration comes down to the GPU. Dell's Lawler says what matters is how much fits on the GPU and how fast it runs, with the rest of the Pro Max system in support. He says the T2 tower, launched on the recording day, grew in capacity for extra drives and cards without much bigger footprint, and runs new Intel processors at a sustained 250 W, which he calls unique in the industry.
• Scarce cloud and data-centre GPUs push experimentation onto the desk. Nvidia's Bali says fine-tuning and running models is now mainstream, even for product marketers like her and Lawler, while data-centre resources are scarce. She cites a data scientist who had to queue in a Slack channel for an instance, and argues developers need a local sandbox to keep up with daily new models and techniques.
• Doubled GPU memory is the new generation's biggest feature. For the RTX Pro Blackwell GPUs, Bali calls the jump from 48 GB to 96 GB per GPU, with up to four in one workstation, the biggest feature, and "a game changer" for running, fine-tuning and serving models locally.

2025-05-02 · open the original ↗

• MCP addresses the integration problem holding back agentic AI. The host says LLMs have always struggled to reach data beyond their training, and custom per-source connectors were brittle and could not scale. Anthropic introduced MCP in November 2024, but only in the past couple of months, he says, has it taken off.
• Adoption has been fast because MCP is open and model-agnostic. He cites early adopters Block, Apollo, Replit and Sourcegraph, and over a thousand community-built MCP servers by February. Any model, Claude, GPT-4 or open source, can use it and anyone can build an integration without permission, positioning it as a "USB or HTTP of AI integration".
• Dynamic discovery lets agents find tools without hard-coding. The host explains that agents detect available MCP servers and their capabilities automatically, so a new server for, say, a CRM is usable immediately. Anthropic provides pre-built servers for Google Drive, Slack and databases. He says MCP complements rather than replaces custom APIs, plug-in systems, LangChain-style frameworks and RAG.
• MCP is the action layer, not a silver bullet. He places it as a standard integration layer for the action component of agents. Its open problems are managing many tool servers, getting models to use tools well, an evolving standard, security and monitoring, and overkill for simple apps. He relays Anthropic's plans for remote servers with OAuth, an official registry, standardised discovery endpoints, streaming and proactive server behaviour.

2025-05-01 · open the original ↗

• Dell collapsed its many PC brands into three tiers. Lawler says the rebrand to Dell, Dell Pro and Dell Pro Max replaces many disparate lines (Precision, OptiPlex, XPS and others) to match how customers shop. Dell Pro Max targets heavy GPU workloads from independent software vendors, such as engineering and Adobe workflows, and was announced at CES and launched at GTC.
• Purpose-built systems bring server power to the desk side. Lawler says the traditional Blackwell-accelerated towers and mobiles remain, but the new part is two system-on-a-chip Grace Blackwell machines purpose-built for data scientists and developers, which he presents as Dell recognising where the market is going.
• Nvidia's software layer makes the hardware a full solution. Bali, who leads Nvidia's AI solutions go-to-market, says its AI software gives a consistent, pre-optimised experience from workstation to server to data centre and back. Developers then need not tune models to hardware themselves; Nvidia and Dell sell the combination as the Dell AI Factory with Nvidia.

2025-04-30 · open the original ↗

• GTC has become a festival-scale event spilling across San Jose. Bali, of Nvidia, describes partners from cloud providers to startups filling the town, many conversations with robots over five days, and people lining up from 6:00 a.m. to get into the keynote, which Jensen Huang preceded by serving food from a Denny's food truck.
• Attendees should arrive with a plan or be overwhelmed. Lawler, at his first GTC, says the volume of learning tracks and booths was overwhelming in a good way, and advises picking a learning track suited to your role beforehand rather than wandering the show floor.

2025-04-29 · open the original ↗

• Scarce cloud compute is pushing AI development back to the desk. Bali, who leads AI solutions go-to-market at Nvidia, recalls a data scientist who had to queue in a Slack channel for an instance, and argues experimentation needs a local sandbox. Nvidia's RTX Pro Blackwell GPUs double memory from 48 to 96 GB per GPU, with up to four in one workstation.
• Nvidia's software layer is meant to remove hand-tuning models to hardware. Bali describes NVIDIA AI Enterprise: NIM microservices package models in containers already optimised for the GPU they land on, so a model can be swapped in with one line of code as Llama versions turn over every couple of months. She cites build.nvidia.com for free prototyping, NeMo for fine-tuning and guardrails, and CUDA libraries such as RAPIDS cuDF (up to 100x on pandas-style preprocessing, no code change) and cuML, announced at GTC (up to 50x on scikit-learn tasks).
• Data-science-only users should choose Dell's Linux GB10 and GB300. Lawler says these Grace Blackwell Dell Pro Max systems ship Linux-only with the Nvidia stack preloaded, while anyone also needing tools like Adobe Creative Cloud, which doesn't run on Linux, should take a traditional Windows Pro Max. Bali adds that the palm-sized GB10 can sit beside a Windows laptop, and two can be daisy-chained over ConnectX.
• The GB300 offers server-class power without server infrastructure. Lawler gives it 784 GB of unified memory (288 GB GPU, 496 GB CPU) and 20,000 TOPS at FP4, against roughly 4,000 for a single 96 GB RTX Pro 6000, and pitches it at mid-market firms with no racks or cooling. Pricing was unreleased at recording; he expects the GB10 near Nvidia's DGX Spark ("3 to 4,000") and the GB300 to cost more but less than a server.
• Both guests forecast AI spreading to non-developers and the physical world. Lawler, not a developer, says he trained his own LoRAs for an animation studio and imagines a RAG system over his mother's recipe cards. Bali, echoing Jensen Huang's GTC keynote, predicts a shift from generative to agentic AI, with agents handing off across a factory, and then physical AI such as driverless cars, adding her personal hope for guardrails and regulation.

2025-04-28 · open the original ↗

• Large language models are here to stay, but every layer around them is in flux. Webber says integration, fine-tuning, agentic systems, pre-training, data-set selection and evaluation will all keep changing, while the core problem of training a model as well as possible and serving it efficiently has not changed and, she expects, never will.
• Knowledge belongs inside the model wherever possible. Drawing on her SageMaker years, Webber argues that pushing knowledge into the network simplifies development, data management and application management, and calls RAG and agentic layers "fluff" next to what the weights hold. She predicts problems first solved at the agentic level will be absorbed into RAG, then pre-training, then the data set itself.
• AWS's accelerator roadmap is only beginning, by her account. Webber says Trainium 3, pre-announced at re:Invent, is on the way, and that customers are "very much just getting started" with what Trainium and Inferentia will deliver.

2025-04-27 · open the original ↗

• Trainium and Inferentia share a core and differ in topology. Webber says both use the same NeuronCore and software stack, so workloads can move between them. Trn1 connects cards in a 4D torus suited to gathering results across cards for the backward pass and optimizer update, while Inferentia lays cards out more simply for sharding a large tensor through a forward pass.
• Inferentia offers more sizes because inference needs vary more. Webber says Inferentia instances come in many combinations of accelerator count and HBM capacity, suited to hosting a 7 billion or 11 billion parameter model, while Trainium comes essentially small and very large: develop on a single Trn1, then scale to as many large instances as you can get.

2025-04-26 · open the original ↗

• AWS builds its own accelerators to give customers hardware choice. Webber frames AWS's position as choice over data sets, models and accelerated hardware, and says customers pick Trainium and Inferentia above all for price-performance and energy efficiency. She claims Trn2 is the most powerful EC2 instance on AWS for AI/ML.
• Annapurna Labs' track record underpins the chips, in her telling. Webber says Amazon acquired the startup in 2015 to rebuild the hypervisor, and its Nitro system physically separates customer data from instance control, which she credits for both AWS's scaling and its security; every modern EC2 instance runs on it.
• Graviton CPUs now take most new AWS compute. Webber cites re:Invent: more than half of new compute coming onto AWS is Annapurna's Arm-based Graviton, flagged by a trailing "g" in instance names such as C6g. Trainium and Inferentia are its third product line, focused on AI/ML.

2025-04-25 · open the original ↗

• ASML alone can build the machines behind leading-edge AI chips. The host says the Dutch firm's latest lithography system weighs 150 tons, is roughly the size of two shipping containers and costs $350 million. No other company can reliably print features below 7 nanometres, and ASML holds over 90% of the market even for mature technologies, supplying TSMC, Samsung and Intel.
• EUV light comes from tin plasma about 40 times hotter than the sun. The host describes the machine firing 50,000 droplets of molten tin into a vacuum, each hit twice by lasers, producing plasma near 220,000°C. The light is bounced off mirrors with imperfections measured in trillionths of a metre onto a chip blueprint. The resulting processors can hold over 100 billion transistors, more than 70 layers and over 100 km of wiring on silicon about 1.5 times a US postage stamp.
• Lithography leadership is measured in decades, not months. The host argues that ASML spent two decades perfecting EUV and money cannot compress that. Its high-NA EUV raises aperture from 0.33 to 0.55 to print 8 nm features, with mirrors doubling in size and growing 10 times heavier. Hyper-NA above 0.75 is next, and wavelengths around 6 nm are seen as a plan B.
• China is squeezing older tools while building its own. Barred by the US from ASML's most advanced equipment, Chinese makers use multi-patterning on older machines to print features two or four times smaller, at a cost in complexity and speed. A state-owned firm is reportedly progressing on a 28 nm-capable tool. The host says matching EUV would mean replicating ASML's supply chain of more than 5,000 specialised suppliers.
• Canon's nanoimprint bet is cheaper but not yet chip-grade. Per the host, Canon's press-like nanoimprint lithography could in theory reach nanometre accuracy at about 40% lower cost. Defects, alignment and speed have so far confined it to uses like smartphone displays and memory chips, where higher defect rates are tolerable.

2025-04-24 · open the original ↗

• NxD hides most of the Neuron SDK's complexity from developers. Webber says the Neuron SDK spans low-level runtime, driver and compiler up to NeuronX Distributed (NxD), the main modelling library. NxD compiles a model and shards a checkpoint such as Llama or PixArt across an instance's accelerators, handling communication, optimizer updates and the forward pass. Prebuilt NxD model packages let users run alignment, supervised fine-tuning or hosting experiments with little exposed complexity.

2025-04-23 · open the original ↗

• A kernel is where the user overrides the compiler on the chip. Webber explains that PyTorch code for Trainium and Inferentia passes through PyTorch XLA into an HLO graph, which AWS's compiler turns into the chip's instruction set. A kernel replaces that generated code with operations the user defines through NKI, the Neuron Kernel Interface.
• Kernel work is the study of compute optimisation for a workload. Kernels range from a hello-world tensor add to a full forward or backward pass or just an MLP projection, Webber says, and writing one means studying data movement, memory and compute utilisation to improve performance on large language models.

2025-04-22 · open the original ↗

• Customers pick AWS's own accelerators mainly for price-performance. Webber, who moved from SageMaker to the Trainium and Inferentia team because she saw accelerator supply and health become "make or break", frames the chips as part of AWS offering hardware choice. She traces them to Annapurna Labs, acquired in 2015, whose Nitro system and Arm-based Graviton CPUs (over half of new AWS compute, per re:Invent) preceded Trainium, and claims Trn2 is AWS's most powerful AI/ML instance.
• The Neuron software path lets PyTorch code reach the chip, with kernels as an override. Webber explains that PyTorch XLA turns a model into an HLO graph that AWS's compiler lowers to the chip's instruction set, while a kernel written in NKI, the Neuron Kernel Interface, replaces compiler output with user-defined operations tuned for data movement and utilisation. NeuronX Distributed (NxD) handles compiling and sharding checkpoints across accelerators, and Trainium also runs under SageMaker Studio and JumpStart.
• Trn2 quadruples Trn1's compute with the same NeuronCore. Webber gives Trn2 eight NeuronCores and four HBM banks per card versus two and two, 96 GB of HBM per card, 16 cards and 1.5 TB per instance. An UltraServer joins four Trn2 instances, 64 cards over two racks, via NeuronLink with at most two hops card to card. Because the core is unchanged, kernels and NxD code move up easily; she recommends Trn1 for models around 70 billion parameters and Trn2 for much larger ones.
• Tensor-parallel degree is a tuning knob with real costs. Webber corrects the host, who took TP for numeric precision: it is the number of NeuronCores sharing one copy of a tensor. Changing it, from TP8 on Trn1 to TP32–128 on Trn2, shifts time spent in collectives such as all-reduce, batch size and step time, and can help or hurt. Trn2's logical NeuronCore setting regroups 128 visible cores into 64 to give each more HBM bandwidth.
• Anthropic is the flagship customer, and academics get $110 million in credits. Webber cites Project Rainier, a very large Trainium cluster built with Anthropic, plus small-language-model startups and Databricks; the host adds Poolside. Build on Trainium offers $110 million in credits to academic researchers, scaling projects on a cluster of up to 40,000 Trn1 cards, and recently paid $25,000 for the fastest Llama implementation in NKI.

2025-04-21 · open the original ↗

• A model without web access may silently invent what a link says. The host recounts giving OpenAI's o1 pro a URL in deep research, at a time it lacked internet access; its reasoning trace said it would assume what the page contained, and he says the output alone would have misled him had he not checked the trace or the site.
• LLMs graded each other erratically but showed no self-preference. Michaelson describes generating hundreds of SAT essay questions, having four LLMs answer them, then having the same four rank the answers. Results were "wildly unstable" between reruns, but he saw no sign of a model favouring its own answers.
• LLM committees are his answer to evaluation bias. Michaelson still finds LLMs effective at judging LLM output, and suggests playing models off one another, for instance letting a cheap model screen answers and calling an expensive one only when an answer looks questionable. He mentions teacher-student distillation from a recent AWS talk in Ireland but says he hasn't looked into its specifics.

2025-04-20 · open the original ↗

• This upload holds a different clip than its title names. The video posted as "The LLM Revolution Threatens SaaS Businesses" is a copy of the 2025-04-19 clip on deploying models: both run 5:56 and their captions are identical. Greg Michaelson's SaaS argument is summarized under the 2025-04-17 clip and episode 879, dated 2025-04-15.

2025-04-19 · open the original ↗

• Handing models to engineers is a risky dependency, not just a bottleneck. Michaelson argues that the developers who recode data scientists' prototypes usually aren't data scientists and can easily introduce mistakes, while data scientists often lack the skills or infrastructure privileges to serialise, dockerise and deploy models themselves.
• Zerve packages environments and models so data scientists can deploy. Michaelson says every Zerve canvas runs in its own reusable, shareable Docker container, ending dependency mismatches for new hires, and every executed block serialises its variables. An API route can point directly at a trained random forest, and deployments run on Lambda-style serverless infrastructure with no long-running services. Where organisations still forbid data scientists from deploying, a handoff system passes deployable software rather than prototypes to other teams.

2025-04-18 · open the original ↗

• DeepSeek set off an AI mania in Chinese markets and enterprises. The host reports that the Manus agent's registration site crashed within hours of its March 6 launch, with codes reportedly resold on the black market. The Hang Seng Tech Index rose more than 40% from mid-January, and hundreds of large firms, Tencent, city governments, hospitals and universities announced plans to integrate DeepSeek's models.
• Chinese investors are funding startups while calling it a bubble. One Hangzhou-based investor, quoted by the host, said there was "no choice" given the weak economy, with a strategy of entering at an A round and exiting at an A+ round a few months later. On March 6 the central government announced a venture fund of a trillion yuan, about $140 billion, for tech investment.
• Big tech is betting cloud AI will offset weak core businesses. Per the host, Alibaba declared human-like AGI its main objective, released a reasoning model it claims matches DeepSeek, and committed around $53 billion over three years to data centres, more than in the past decade, from a 36% share of China's cloud market. Baidu's cloud revenue has risen.
• The boom is driving a server and hardware spending surge. Demand for AI servers has soared since early February, including all-in-one on-premises units favoured by state-owned firms. A server maker founded by former Huawei staff is up about 140% in 2025, and Jefferies analysts estimate server makers may spend over 1.4 trillion yuan, about $200 billion, over two years.
• Chip access is the boom's chief vulnerability. Morningstar's Kai Wang argues DeepSeek won't fundamentally change most rally beneficiaries, and the host notes an earlier rally faded. Chinese firms still buy Nvidia's weaker H20, local designers such as Cambricon and Huawei lag, and SMIC is capacity-constrained. The host warns that reported US plans to restrict even the H20 could push costs back up and end the euphoria, while welcoming the open-sourcing DeepSeek leads.

2025-04-17 · open the original ↗

• Zerve pitches itself as a full-stack environment for the data science life cycle. Michaelson says that apart from hardware and data warehousing, which it connects to, Zerve covers data connection, exploration, visualisation, reports, GPU model training, self-built or SageMaker deployment and scheduled jobs. SQL, R and Python users work side by side, syncing to GitHub or Bitbucket so engineers can stay in VS Code.
• LLMs have made building so cheap that bespoke SaaS faces an existential crisis. Michaelson argues the old build-versus-buy trade-off has shifted sharply toward building, threatening SaaS vendors that sell custom use-case solutions "and then charging a mint for them", with knock-on effects for venture capital. He doesn't predict what the space will look like in a few years.

2025-04-16 · open the original ↗

• Serverless blocks make parallel execution free of extra code or cost. Michaelson says Zerve's new "fleet" feature, launching the week of recording, fans code out across serverless compute, so a thousand slow LLM calls can run at once instead of in series. Because each block in Zerve's DAG already spins up its own serverless compute, parallelising needs no pool management and, he says, costs no more since total compute is the same.
• A shared DAG canvas does for notebooks what Google Docs did for documents. Code lives in the graph's nodes and data frames flow along its edges, with forks and merges, Michaelson explains, so several people can run Python, R and SQL blocks in one canvas simultaneously. He contrasts Jupyter, where metadata changes cause merge conflicts and "final final" file names; the host supplies the Google Docs analogy.
• Low-code tools hit a ceiling, so value comes from people who code. Michaelson, who spent about seven years at DataRobot, says the emperor of 2010s low-code/no-code "has no clothes", since complex problems exceed such tools, which is why DataRobot and Databricks added notebooks. Zerve, he says, remains code-first, though LLMs have turned coding into describing what you want and then tweaking the result. The host relays Zerve's claim of cutting cycle times by up to nine times; the clip has no evidence for the figure.

2025-04-15 · open the original ↗

• Cheap in-house building with LLMs threatens many SaaS vendors. Michaelson argues the build-versus-buy calculation has shifted so far that firms selling custom use-case software face an existential crisis, though nobody will rebuild something like Salesforce, and he separates "dime a dozen" generative AI bolt-ons from deeply integrated agents. The host sets this against Klarna's and Satya Nadella's similar views and Zoho's CEO's counter-argument that AI will spawn new vertical SaaS, and notes a friend who cut a $20-a-month ChatGPT Plus cost to $2 by calling the API through his own interface.
• Zerve's serverless blocks make parallelism and collaboration nearly free. Michaelson explains that code lives in the nodes of a DAG canvas, with data frames passing along its edges, and each block spins up its own serverless compute. That lets several people run Python, R and SQL in one canvas at once, and lets the new "fleet" feature fan out, say, a thousand slow LLM calls with no extra code and, he says, no extra cost.
• Zerve lets companies choose where prompts go to protect their IP. Michaelson says OpenAI's models are the best in his experience, though he distrusts LLM benchmarks, and sending prompts there is unavoidable with them. Zerve also offers AWS Bedrock and self-hosted Hugging Face models, which can be fine-tuned, and Zerve itself is self-hosted. Beyond prompt-driven GenAI blocks, its AI assistant is an agent that plans, creates and connects blocks from a natural-language request.
• Data scientists should ship software, not prototypes for engineers to recode. Michaelson calls the handoff to developers a risky dependency, since they may misunderstand the model. Zerve gives each canvas a Docker container and serialises every block's variables, so an API route can point straight at a trained random forest and deploy on Lambda-style serverless compute, with a handoff path where data scientists aren't allowed to deploy.
• LLM committees are his answer to bias and error in AI judgment. In Michaelson's test, four LLMs answered and then ranked responses to about a hundred SAT essay questions; rankings were "wildly unstable" across reruns, but no model favoured its own answers. He suggests cheap models screening answers and escalating doubtful ones to expensive models, and warns that a model that gets a problem badly wrong first time is rarely correctable.

2025-04-14 · open the original ↗

• Dell estimates its toolkit cuts NPU app delivery from six months to six weeks. Gupta says that by Dell's estimate a typical, not very complex AI PC app such as a RAG chatbot takes about six months from discovery to fleet deployment, covering model and device selection, runtime compatibility, on-device model operations and manual deployment. Dell Pro AI Studio, in early access at recording and due fully mid-2025, would bring that under six weeks, about a 75% reduction, leaving developers to spend time on picking the use case and validating output accuracy.

2025-04-13 · open the original ↗

• Today's NPUs comfortably run quantised models of about 8 billion parameters. Gupta puts acceptable on-device output at around 15 to 20 tokens per second, which he says rules out going much beyond an 8 or 9 billion parameter model with quantisation. NPUs passing 40 TOPS and smaller LLMs growing more capable are, in his view, the inflection that makes local inference practical.
• Dell-validated models end trial-and-error pairing of models and machines. Gupta describes developers otherwise iterating through 14 and 30 billion parameter models and memory upgrades only to land at two tokens per second, a deterrent to building for the NPU. Dell Pro AI Studio, a developer and IT toolkit rather than a preinstalled app, offers open, permissively licensed language, speech and vision models tested against Dell systems across silicon vendors, published on the Dell Enterprise Hub on Hugging Face.
• Enterprise readiness and on-device middleware are what local-chat demos lack. Gupta says enterprises need attestable models, control of data in and out, and lifecycle management of permissions, updates, deprecations and telemetry, which a Dell management portal provides. On-device middleware detects the silicon, pulls the intended model from a registry, loads, unloads and verifies it, replacing hundreds of lines of code developers would otherwise write and rebuild with each model change.

2025-04-12 · open the original ↗

• Build anywhere, but integrate and test AI PC apps on Windows. Gupta argues that because NPU inference apps will mostly be consumed by Windows users, developers can do much of their work in an IDE of choice on Unix-based machines, but must integrate the model into the app, set up the local host and verify it runs on a Windows device, since that is where the app will run.

2025-04-11 · open the original ↗

• AGI needs planning over an unlimited horizon, which LLMs lack. Burkov, in an episode 867 clip, contrasts crows and monkeys that plan a day or two ahead with humans who plan for generations, and says LLMs are reactive: an "agent" has only the agency its system prompt gives it. He doesn't hazard a timeline, likening AGI to teleportation. The host suggests it may need brain-like modules such as a prefrontal cortex rather than a scaled-up transformer.
• Models can't tell what they don't know, and fine-tuning hacks don't fix it. Burkov says o3-mini's visible reasoning admitted it knew nothing of his new book, released after its cutoff, yet its answer praised the book as if fact. He says Qwen models fine-tuned to say "I don't know this person" denied knowing him, then named him as author of his first book.
• Doubling down on embodied experience is the human edge over AI. In an episode 873 clip, a digital-twin entrepreneur named Natalie argues AI is part of our cognitive evolution as language is, but not all of us. Embodied, serendipitous experience, like the host's childhood browsing of dictionaries, builds insight AI can't replicate, so people should collaborate with AI and use it to free themselves to be more human.
• The AI stack looks different to executives and developers. Richmond, a MongoDB software engineer, says in an episode 871 clip that executives see three layers: application (such as Cursor), tooling (such as MongoDB) and compute (such as Nvidia). Developers need more, starting with programming language, since many libraries exist only in Python or JavaScript, then model provider and database.
• Google's tuning playbook tries to make neural-network training systematic. Varun, a former Google engineer and first author, says in an episode 869 clip that the playbook moves training from "alchemy" toward chemistry. Architecture-agnostic, it helps practitioners choose the next experiment given a compute budget. It was released as a Markdown file on GitHub so others could fork it, and had 28,000 stars at recording, per the host.

2025-04-10 · open the original ↗

• Local accelerators give low latency for persistent real-time work. Gupta's own mnemonic starts with "accelerated": an on-device NPU suits latency-sensitive, always-on tasks such as translation, transcription and captioning.
• On-device AI can learn an individual user's style. For "individualized", Gupta cites a healthcare customer auto-generating ER physicians' reports from their diagnoses; a physician reported that as the model saw his edits it began to sound like him, freeing time for patients.
• Data never has to leave the device. For "private", Gupta argues local models avoid sending confidential or PII-laden data even to a private cloud tenant.
• Offloading inference to PCs cuts data-centre and per-token costs. For "cost", Gupta says a financial-services customer found 15% of its data-centre compute going to developers' code generation and completion, work he believes NPUs can take over without H100s, and cloud inference bills per token. He concedes cloud inference costs are falling, but argues for "the right engine for the right workload". Edge vision cases include real-time factory defect detection and an insurer classifying damage photos into rough cost sizes to get adjusters "maybe 70% of the way there".
• Dell works with chip partners so models run across NPU silicon. Asked about chips from Intel, Qualcomm and AMD, Gupta says the NPU architecture is so new that Dell is preparing models with its silicon partners. This lets developers target different chips without starting over, one reason Dell built Dell Pro AI Studio.

2025-04-09 · open the original ↗

• NPUs are power-efficient matrix-math engines new to ordinary PCs. Gupta says NPUs reached the market about a year before recording, first in Intel's Meteor Lake chipset, joining the CPUs and integrated or discrete GPUs PCs already had. Being purpose-built, almost hard-coded, for the matrix multiplication and addition underlying AI, they use far less power than running models on a CPU or GPU, which would drain a laptop battery.
• For now NPUs are for inference, while GPUs still build the models. Gupta expects data scientists and developers to keep using GPU workstations for training and fine-tuning. NPUs will serve the knowledge workers who consume the results, from assistants answering questions over a knowledge base to AI embedded in workflows and, he foresees, multi-agent systems, without tanking battery life.

2025-04-08 · open the original ↗

• NPUs are inference engines for end users, not training hardware. Gupta, of Dell, says NPUs arrived about a year before recording with Intel's Meteor Lake and are near-hard-coded for matrix math, so they run AI far more power-efficiently than a PC's CPU or GPU. GPUs remain for training and fine-tuning. He puts acceptable output at 15 to 20 tokens per second, which limits today's NPUs to quantised models of about 8 or 9 billion parameters, now capable enough to make local inference practical.
• His "AIPC" mnemonic makes the case for on-device AI. Gupta's letters stand for accelerated (low latency for translation or captioning), individualized (an ER report generator that learned a physician's style), private (data never leaves the device) and cost. On cost, a financial-services customer found 15% of its data-centre compute going to developers' code completion, which he thinks NPUs could absorb. He concedes cloud inference is getting cheaper and that not every use case belongs on the device.
• Dell says its toolkit cuts NPU app delivery from six months to six weeks. Gupta describes Dell Pro AI Studio, in early access at recording and due mid-2025, as three parts. Validated models on the Dell Enterprise Hub on Hugging Face end trial-and-error pairing of models and machines; a management portal adds attestation, permissions, updates and telemetry; on-device middleware detects the silicon and loads the right model. By Dell's estimate, a typical app such as a RAG chatbot goes from discovery to fleet deployment about 75% faster.
• Build on workstations, scale on servers, deploy to AI PCs. Gupta advises embedding data scientists with the people whose problem needs solving, then moving production inference to AI PCs if H100-class hardware isn't needed. His example is a healthcare customer that trained a custom vision transformer on its own radiology images and reports physician-level diagnostic accuracy. It now runs in a HIPAA-compliant data centre, and the customer wants to bring it to AI PCs for doctors worldwide.
• On-device agents await small reasoning models; hybrid compute comes sooner. Gupta forecasts agents acting on users' behalf within set parameters, such as buying concert tickets on approval, but says reasoning models must first shrink enough to run locally, which "might take some time". He also expects workloads to shift seamlessly between device and cloud as compute needs demand.

2025-04-07 · open the original ↗

• Merck's display dominance plateaued, so it moved into semiconductors. Beckmann, who heads Merck KGaA's 8,000-employee electronics business, says its late-1960s and 1970s liquid-crystal research took decades to commercialise. At peak, roughly 2005 to 2010, it held 80% of the liquid-crystal materials market. Recognising that every innovation ramps then plateaus, it entered semiconductors around 2014 and built a leading portfolio through acquisitions including AZ, Sigma-Aldrich's high-tech business, Versum and Intermolecular.
• Incumbents' worst innovation mistake is assuming yesterday's success lasts. Beckmann argues that innovating inside an established company is far harder than in a startup because "you are in your own way", and calls believing that what worked yesterday will work tomorrow probably the most fatal mistake in business. He traces Merck from its 1668 founding after the Thirty Years' War to what he says was industry's first quality promise to customers around 1850.

2025-04-06 · open the original ↗

• High-bandwidth memory depends on hundreds of specialised materials. Beckmann says HBM needs the best possible DRAM performance, and DRAM capacitors rely on high-k materials where he says Merck leads. Spin-on dielectrics are among many promising materials, but with hundreds or thousands of precursors drawn from 80% of the periodic table, he declines to single out one as most important.
• Light-based data transfer promises speed and energy gains. Beckmann says photons move data faster than electrons and without the same heat, and that Merck reorganised its electronics sector the year before recording to create an optronics unit built on its display-era expertise in generating and manipulating light.

2025-04-05 · open the original ↗

• Heterogeneous integration scales performance beyond shrinking transistors. Beckmann describes it as the step between front-end transistor making and back-end packaging: combining several dies into one system, stacking dies into memory stacks or placing memory beside a GPU, as in CoWoS-style structures, to shorten data paths. It requires far smaller structures, better heat removal and power optimisation, and front-end-like precision, which opens room for materials and metrology innovation.
• Merck uses AI to skip unnecessary materials experiments. Beckmann says its "materials intelligence" approach applies AI across millions of candidate options to predict how a material will behave in a customer's setup, rather than improving properties sequentially by experiment.

2025-04-04 · open the original ↗

• smolagents keeps agent logic small enough to read and tweak. The host describes Hugging Face's new Python library as giving any LLM the ability to use tools, call APIs and reason over multiple steps. Its core logic fits in a few thousand lines with minimal abstraction over raw Python, and it succeeds the transformers.agents module, which it will eventually replace.
• Its agents act by writing Python code, run in a sandbox. Rather than expressing actions in JSON or text, smolagents' code agents generate Python that is then executed, which the host says gives them Python's full flexibility, with sandboxed execution as the safeguard. A conventional tool-calling agent is also offered.
• Tools are plain decorated functions, shareable on the Hub, with any model. Any Python function becomes a tool with an @tool decorator, type hints and a docstring, and tools or whole agents can be shared and loaded from the Hugging Face Hub, "like an app store". Models can be open ones via Transformers or the inference API, proprietary ones from OpenAI or Anthropic via LiteLLM, or local ones.
• It lowers the bar for agents in business, data science and research. The host lists research question-answering, as in Hugging Face's demo of a leopard running across a Paris landmark, one-line CLI trip planning, browser automation via its web agent, and on-the-fly pandas analysis. He argues firms can prototype agents in a few lines and keep data in-house with local models, and researchers can benchmark small open models against large proprietary ones. He states the episode is not sponsored by Hugging Face.
• Heavier frameworks beat it for scale, data and multi-agent work. By the host's comparison, LangChain surpasses smolagents in scalability, memory management and integrations, and LlamaIndex in large-data retrieval. Pydantic AI offers type-safe validated outputs, while Microsoft AutoGen and CrewAI are built for coordinated multi-agent systems, where smolagents focuses mainly on single agents.

2025-04-03 · open the original ↗

• A chip takes about 1,400 repeated steps from blank wafer. Beckmann explains that leading chipmakers build transistors and power and data lines on 300 mm silicon wafers, the size of a vinyl record, cycling through lithography, patterning, deposition, cleaning and planarisation until the wafer is cut into centimetre-scale dies.
• Chipmaking draws on about 80% of the usable periodic table. Beyond photoresists for lithography, Beckmann says the specialty materials for etching, deposition and planarisation span about 80% of the non-radioactive periodic table. His examples mostly concern logic chips such as CPUs and GPUs, which he calls the main driver of compute, as distinct from memory.
• His route to leading Merck's electronics arm ran through IT and HR. Beckmann studied computer science with a semiconductor focus and worked on chip design in the late 1980s, then spent years in Merck's corporate IT, ran a country sales organisation, and served as board member for HR during a period of transformation before being asked to head electronic materials.

2025-04-02 · open the original ↗

• Neuromorphic chips may pay off first in inference, within about five years. Beckmann argues neuromorphic architectures could matter even more for cost-efficient, scalable inference on everyday devices than for training, given that data transfer is a key optimisation target. He calls their use in general compute "a five years plus question rather than a 10 years question", but says it probably won't reach human-brain efficiency; the light-bulb-powered brain comparison is the host's.
• Data-centre AI already scales on several fronts at once. Beckmann says chips such as Nvidia's Hopper and Blackwell combine transistor shrinkage via EUV lithography and gate-all-around nanosheet designs with CoWoS-style integration that places memory beside compute. The same latest technologies are reaching smartphones, laptops, cars and production systems.

2025-04-01 · open the original ↗

• Chip performance now scales along three dimensions, not just shrinkage. Beckmann, CEO of Merck KGaA's electronics business, distinguishes "more Moore" (smaller transistors via EUV and gate-all-around nanosheets), "more than Moore" (heterogeneous integration such as CoWoS, placing memory stacks beside GPUs as in Hopper and Blackwell) and post-von Neumann architectures. He expects neuromorphic computing in general compute in "five years plus" rather than 10, though not at human-brain efficiency. Superconducting quantum approaches need near-zero-kelvin cooling and remain a tiny market; Merck's Intermolecular facility in San Jose is exploring qubit materials with partners.
• Co-optimising materials, delivery and inspection is meant to compress years-long cycles. Beckmann says chipmakers' yield problems are solved fastest by integrating chemistry, physics, electrical testing and chemical delivery systems. That is the logic for acquiring UnitySC, whose high-resolution visual inspection checks 2.5D and 3D structures such as through-silicon vias. Merck calls its use of AI, including on metrology data, to cut unnecessary materials experiments "AI for AI".
• AI drives leading-edge chips while the rest of the industry stays in a cyclical slump. Beckmann says volume still depends on smartphone and PC replacement, which remains low as people keep COVID-era devices into a fourth or fifth year, and AI data-centre demand "doesn't bail everything else out". Because fabs run either at full capacity or idle, the industry is inherently cyclical; after semiconductor-materials growth above 20%, he says Merck stays humble, knowing shrinking years follow.
• Good AI regulation protects data contributors without slowing permits. Beckmann argues frameworks should ensure those who make data available get returns when it feeds LLMs, since that stability helps investment. But he says that needing about a year to get a permit to make a new material in Darmstadt hurts innovation, and that Europe is not leading on approval speed.
• Every innovation plateaus, so incumbents must switch gears deliberately. Beckmann recounts that Merck's liquid-crystal research from the late 1960s led to an 80% share of liquid-crystal materials at the flat-panel peak around 2005 to 2010. It then moved into semiconductors from 2014 through acquisitions. He calls assuming yesterday's success will last "probably the most fatal mistake you can do in business".

2025-03-31 · open the original ↗

• Experts can rent out vetted AI twins of themselves. Monbiot says she now advises a startup (spelled on air W-I-Z-L-Y) that admits experts selectively, lets them train a twin on knowledge they control, use it to generate leads, and monetize it in consulting calls once they trust it to represent them.
• A twin's value is augmented skill, not just bought time. Because the twin is cheaper, always available and speaks any language, Monbiot argues it can serve clients abroad at a price point they can afford, which she calls the "10x" of the idea.
• A CEO's twin can carry company culture across an organization. Monbiot describes building one for a healthcare CEO, first inside Slack, where employees can consult it on his values, have it interject, and get feedback on work, including staff at a company recently acquired; voice and other modalities are to follow. The host flags liability as an open question when people act on the twin's advice.

2025-03-30 · open the original ↗

• Each step from books to Google to LLMs removes cognitive work. The host argues that looking things up in a dictionary or encyclopedia exposed him to incidental information that built a web of connections and aided memory, while a search page surfaces ads and an LLM output does the comparing for you.
• Humans should double down on embodied experience to stay competitive with AI. Monbiot holds that AI is part of human cognitive evolution the way language is, but not all of us: being embodied and embedded in the world, noticing things and making real-world connections is the human advantage AI lacks. Her prescription is to use AI to solve big problems and free ourselves "to be more human".

2025-03-29 · open the original ↗

• Outsourcing whole cognitive processes to LLMs makes us dumber, unlike Google. Monbiot answers commenters who compare the worry to early fears about Google: search returns facts that feed into your own reasoning, whereas an LLM can tempt you to hand over the entire process without reviewing or challenging the result. She expects this to change how children are educated and how adults reskill to avoid that trajectory.

2025-03-28 · open the original ↗

• Baseball analytics has gone from Moneyball experiment to AI arms race. The host says every MLB team now employs data scientists, and that Statcast, introduced in 2015 with high-speed cameras and radar in every ballpark, generates seven terabytes of data per game, far more than human analysts can sift.
• Machine-learning scouting finds players that traditional stats undervalue. Models ingest college and minor-league stats, biomechanics and video, and weight metrics such as exit velocity, launch angle and spin rate; the host's example is a prospect with a low batting average who consistently hits the ball hard. Teams also use the analysis to tailor drills, such as for a hitter who struggles with high fastballs.
• In-game decisions now lean on simulated matchups and positioning models. Managers consult win-probability tables and pitcher-versus-batter predictions on tablets in the dugout, and data-driven defensive shifts proved effective enough that MLB introduced rules limiting extreme shifts. The host stresses the tools augment managers rather than replace them.
• AI is reaching fans and officiating as well. Broadcasts show live win-probability graphs and catch probability on replays, and the host reports that during 2025 spring training MLB let players challenge human umpires' ball-strike calls to an automated system, which drew mostly positive reviews.
• Any traditional industry can be transformed by pairing domain expertise with AI. The host's lesson is that if America's oldest major sport can reinvent itself, finance, healthcare, manufacturing and retail can too; early adopters gain an edge until others catch up. He forecasts AI won't replace product managers, physicians or supply-chain planners for the foreseeable future, but will arm them with better information.

2025-03-27 · open the original ↗

• Step-by-step reasoning models could scale to many Nobel-level minds. The host, prompted by OpenAI's o1 release, argues synthetic data could broaden reasoning beyond math, chemistry and code, and forecasts that within "a matter of years" a data center could hold a million Nobel-prize-level machines, the "powerful AI" of Anthropic CEO Dario Amodei's essay.
• Relying on LLMs for cognitive work makes people dumber. Monbiot cites studies showing users' cognitive capabilities decline when LLMs do their thinking, and warns of a "great intelligence divide" in which a few build ever more powerful AI while everyday users degrade. Her focus is on how AI can augment human agency rather than on alignment.
• AI's real payoff is freed capacity for ingenuity, not a modest speed-up. She calls the roughly 15% average performance gain for white-collar workers found by Ethan Mollick and others uncompelling, and prefers Andrew Ng's framing of the 10x engineer as exponentially more effective because AI frees time and mental capacity.
• Used well, AI expects more of people, not less. Monbiot cites the former CEO of Drift building an AI-agent-first customer-success startup whose first hire after engineers was a lawyer who also oversees HR. She argues people must think laterally and drop ego: engineers whose work is democratized, and non-technical people who can no longer claim "I'm not a coder".
• The "anyone can now code" narrative isn't fully true yet. Monbiot, a linguist by training, says she got halfway through a Python course, tried building, and had to go back and learn more. The host argues the best learning pairs a specific goal with just enough fundamentals, with the balance differing by person.

2025-03-26 · open the original ↗

• Small changes to a technology like language can reshape civilization. Monbiot cites a scholar's argument that the ancient Greeks' addition of vowels made expression more fluid and helped language spread, contributing to the birth of democracy, an impact only visible in hindsight. She argues AI's effect on humanity is similarly impossible to predict yet but probably huge.
• Advanced language arrived long after brains capable of it, the host says. He puts sophisticated language at only tens of thousands of years old while saying Homo sapiens have been genetically "pretty much the same for millions of years", and tells listeners to double-check that fact rather than quote him.

2025-03-25 · open the original ↗

• Humans should be AI's meaning makers, not lemmings off a cliff. Monbiot traces her recent writing to the book God, Human, Animal, Machine, which she recommended on her previous appearance (episode 823), and to an op-ed in the Polish magazine Przekrój asking whether we are ceding humanity to machines. Her answer: humans created the opportunity and the mess, and are good at cooperating to fix messes.
• AI is part of human cognitive evolution, as language was. Drawing on John Vervaeke's lecture series The Meaning Crisis, she calls language a "psychotechnology" that enabled self-reflection and cites his argument that Greek vowels helped give birth to democracy. AI is part of us the way language is, but not all of us, because humans are embodied and embedded in the world.
• Leaning on LLMs risks a "great intelligence divide". Monbiot cites studies that users' cognitive capabilities decline when LLMs do their thinking, and dismisses the roughly 15% average gain for white-collar workers as uncompelling. She favours Andrew Ng's 10x-engineer framing, and the example of an AI-native startup whose first post-engineer hire was a lawyer also running HR, as signs AI should demand more of people.
• LLM drafting drags writing toward the mean. Monbiot says the more iterations she does with an LLM, the more her "aha" insights get averaged out and she has to reinsert them, so she is unsure it saves her time. She and the host agree the states worth optimizing for are flow and peace; the host argues free, ad-funded platforms are built to keep users in a frenetic, fearful state instead.
• Everyone will have AI twins working on their behalf, she predicts. Monbiot's "virtual human economy" thesis drives her advisory work for a startup offering vetted experts AI twins that can earn consulting income in any language, and a custom twin of a healthcare CEO deployed in Slack to carry company culture. The host describes his own early idea for an embodied AI that would steer him back toward flow and peace.

2025-03-24 · open the original ↗

• Build AI strategy on pillars and objective truths, then derive tactics. Alake, who leads and mentors AI teams at MongoDB, calls this the year of "show me the money" for AI and says his method is first-principles: identify the pillars around a persona, then the things within each pillar that will not change short of a platform shift. His example for a developer-focused strategy is GitHub as a pillar, asking what developers actually do there, then forming convictions, assumptions and tactics from that.

2025-03-23 · open the original ↗

• This is the "multi" era, not just the year of agents. Alake calls 2025 the "multi" era: agents went mainstream this year, so leaders should look further ahead: to multi-agent architectures, multimodal embeddings covering text, images, video and audio in one space, multilingual embeddings, multi-step reasoning, and multiple retrieval mechanisms, since "vector search is not all you need".
• Voice agents will be widely built but won't dominate, he predicts. Alake expects everyone to experiment with voice interfaces given AI's experimental nature, while holding a strong opinion that voice will not become the main way people engage with agents.
• Memory management will become a central agent-design problem. He recounts that Andrew Ng's first question to him, in April of last year, was how to model memory in agentic systems, and argues agents will need several forms of memory and recall, as human brains do.

2025-03-22 · open the original ↗

• The AI stack is described differently to executives and to developers. Alake, whose blog post on AI stacks the host says was the top Google result for the term at recording, gives executives three layers: application (e.g. Cursor), tooling (e.g. MongoDB) and compute (e.g. Nvidia). For developers he breaks it into more layers, including programming language, model provider and database.
• Choice of programming language constrains the rest of the AI stack. Alake stresses that many libraries further up the stack exist only in Python or only in JavaScript, though some are evolving to support both, so the language decision comes early and matters.

2025-03-21 · open the original ↗

• This upload holds a different episode than its title names. The video posted as 872 is a copy of episode 870 on OpenAI's Deep Research: both run 17:10, their captions are identical, and neither mentions the Majorana 1 chip. Episode 870's summary, dated 2025-03-14, covers what the video says.

2025-03-20 · open the original ↗

• MongoDB Atlas exists because no developer wants to be a database administrator. Alake describes Atlas as a managed, distributed MongoDB running across AWS, Azure and Google Cloud in multiple regions, bundled with streaming and dedicated search nodes.
• Vector spaces place similar meanings close together. The host explains embeddings from Tomas Mikolov's word2vec, around 2011–2012, which learns a word's meaning from surrounding words without labels; in a trained space, spelling variants coincide, NoSQL sits near MongoDB, and dimensions can run to hundreds or thousands. Documents embedded the same way let RAG systems retrieve relevant items from, say, a million documents by a query's meaning.
• MongoDB folded search and vector search into one database. Alake says customers had been bolting search engines, extensions or separate databases onto MongoDB, so it added Atlas Search and then vector search, allowing lexical, vector and hybrid search in a single database where he has seen some teams run two or three.

2025-03-19 · open the original ↗

• MongoDB's document model mirrors how developers think. Alake contrasts relational databases' tabular rows and columns with MongoDB's key-value, JSON-like documents, and says seeing data represented the same way in the application and data layers made him a full-stack MEAN developer (MongoDB, Express, Angular, Node.js) after he had disliked programming.
• Flexible schemas suit fast-moving AI experimentation. The host says his startups chose MongoDB around 2014 because pages' data needs were uncertain; Alake argues AI work is the same, with schemas evolving during experimentation and lockable once stable in production. The host's example is storing extra output fields from a Llama model that the OpenAI API lacks, without a schema migration.
• JSON unifies the AI stack from LLM to database. Alake argues that with LLMs now offering JSON mode and structured output, JSON runs through the application layer, the data layer and the model, and that avoiding schema migrations matters because developer speed is a competitive advantage.

2025-03-18 · open the original ↗

• A flexible document schema suits AI's constant experimentation. Alake, a staff developer advocate for AI/ML at MongoDB, argues the JSON-like document model lets the data layer evolve while requirements are unclear and be locked once in production, avoiding schema migrations; with LLMs offering JSON mode and structured output, JSON now unifies application, data and model. The host, who says MongoDB does not sponsor the episode, gives the example of storing an extra output field from a Llama model that the OpenAI API lacks.
• One database can handle operational, lexical, vector and hybrid search. Alake says customers had bolted search engines, extensions or separate vector databases onto MongoDB, sometimes running two or three databases, so MongoDB built Atlas Search and then vector search into the managed, multi-cloud Atlas service. The host adds that a single vendor avoids version mismatches between linked systems.
• Store an agent's tools in the database; retrieve only a few. Alake's open-source library Memorizz implements a "MongoDB as a toolbox" pattern: since OpenAI's guidance is to put only 10 to 20 tool schemas in context, store perhaps a thousand tools with embeddings of their docstrings, then fetch two or three before each LLM call. He jokes that stuffing a million-token context window on every call depends on "how deep your pocket goes".
• The "multi" era is here, and memory is the next frontier. Calling 2025 the "multi" era, Alake expects, beyond single agents, multi-agent architectures, multimodal and multilingual embeddings, multi-step reasoning and multiple retrieval mechanisms, since "vector search is not all you need". He predicts voice agents will be widely tried but not dominant, and that memory management, as pursued by MemGPT's maker Letta and the HippoRAG paper, will become central.
• Ground AI strategy in objective truths; test tactics in the market. Alake's "arena framework" pictures a coliseum whose pillars are segments of objective truth (for developers, GitHub and what they always do there), with tactics battling it out in the market. For teams he prescribes being aggressive, centralized, coordinated, resourceful, efficient and data-driven.

2025-03-17 · open the original ↗

• Extrapolate current curves a little rather than predict AGI. Godbole, recently retired from Google, says his post "From Knowledge to Wisdom" starts from observable trends: each year models get much better while inference cost for the same functionality falls by orders of magnitude, context windows grow and latency improves, as with Tesla's Roadster preceding cheaper models. He notes there is no consensus definition of AGI.
• AI coding tools will produce "full-stack employees". Because tools such as Cursor, Lovable, Replit and Copilot make software accessible to non-engineers, Godbole expects a PM to arrive with generated mocks, a prototype and early user research before an engineer spends a second on it. He suggests roles are bundles of skills that may be unbundled and dynamically rebundled, leaving the shape of future organizations unclear.
• Cheap reproduction could make software resemble the music industry. If an undifferentiated SaaS product can be cheaply copied, Godbole reasons through a chain of "ifs" that most offerings face near-perfect competition while a few highly differentiated winners take outsized returns. He expects the effects to be discontinuous, such as small teams simply not hiring a PM.
• When knowledge is commoditized, wisdom becomes the differentiator. Godbole argues good ideas still come from humans, and that cultivating wisdom resembles cultivating curiosity, open-mindedness and adaptability. His aim is not more agentic AI but using AI to make himself more agentic, self-authored and pro-social, which he says also improved how he wields models, since language makes human assumptions the models do not share.

2025-03-16 · open the original ↗

• LLMs have turned much of machine learning into prompting plus evals. Godbole, a core member of Google's Gemini team until he recently left, calls LLMs an "arbitrage of ML talent": where adding intelligence to a product once meant PyTorch or JAX and data collection, it now means articulating the behavior you want in clear English and measuring it with good evals, a very different skill set.
• The mental model in his prompting playbook outlasts its prescriptions. Godbole says the second half of the playbook he published before leaving Google, examples of good and bad prompts, reflects today's models, since prompting o1 or Gemini Flash Thinking feels different from GPT-4o or Gemini 1.5. The first half, on pre-training versus post-training, he considers more future-proof.
• Pre-training learns every "cinematic universe"; post-training picks one. He argues that for an LLM there is no objective fact: "Bilbo Baggins is a hobbit" is true only in Tolkien's universe, and web text approximates the union of all such universes. Post-training narrows the model to an assistant-and-user universe, so post-trained models are statistical models role-playing as human raters role-playing as AI assistants.
• Prompt as if briefing an all-knowing stranger who doesn't know you. Godbole says complaints about why a prompt is failing usually work better when pasted into the prompt, and that models respond best to extremely explicit, low-context communication. He recommends the book Nonviolent Communication, learned in couples therapy, because being explicit about the observable behavior you want works on people and models alike.

2025-03-15 · open the original ↗

• Google bet on ML code completion before ChatGPT made it obvious. The host points to a July 2022 Google Research post, "ML-Enhanced Code Completion Improves Developer Productivity", noting that at the time such tools might have seemed more trouble than their errors were worth. Godbole, who moved from medical imaging to this work, says early results were promising and that steadily improving hardware and model size made engineering productivity a natural use case for sequence-to-sequence models.

2025-03-14 · open the original ↗

• OpenAI's Deep Research does hours or days of literature review in minutes. The host, who notes OpenAI does not sponsor him, explains that it breaks a query into sub-tasks, searches the web for each, and iteratively synthesizes a cited report, pivoting as it learns. OpenAI trained it with end-to-end reinforcement learning on browsing and reasoning tasks, and it can read uploaded files, plot with Python, embed images and cite specific passages.
• Deep Research jumped far ahead on Humanity's Last Exam. On the 3,000-question benchmark spanning over 100 subjects, the host's table at recording shows GPT-4o at 3%, Grok 2 and Claude 3.5 Sonnet around 4%, Gemini Thinking about 6%, o1 and DeepSeek R1 around 9%, o3-mini (high) about 13%, and Deep Research at 27%. He predicts that, as with past coding and math benchmarks, the exam won't take long to be conquered once AI gains traction.
• In the host's use, it asked sharp questions and saved hours. Drafting the syllabus, title and abstract for his four-hour agentic AI workshop at ODSC East, he supplied partial modules and past abstracts; it asked about audience level, hands-on coding and tone, then spent three minutes across eight sources, including his own website, to produce a draft he only lightly edited.
• Google's and Perplexity's versions trade depth for other strengths, the host judges. He says Google's Gemini-based tool shows a user-guided research plan first and is reliable but less nuanced, while Perplexity's is fast and free but shallower; for mission-critical analysis he gives OpenAI's the edge despite its cost.
• The catches are cost and occasional overconfidence. The host pays $200 a month for Pro, which includes 100 Deep Research queries a month, and warns it can hallucinate or present rumor as fact, though he has caught neither and says OpenAI reports markedly lower hallucination rates. He forecasts efficiencies such as smaller models like o3-mini will bring more queries to all paid users and eventually free access.

2025-03-13 · open the original ↗

• The Deep Learning Tuning Playbook aims to turn alchemy into chemistry. Godbole, its first author, says it pooled his Google Brain co-authors' decades of experience into a systematic, architecture-agnostic process: set up a good initial experiment, then decide each next experiment from the data collected, given a compute budget and constraints. He credits the co-authors as the real brains and calls it their "two cents", not the final answer.
• Releasing it as a forkable markdown file made it a community resource. Rather than a PDF on arXiv, the team published it on GitHub under a permissive license to invite forks and pull requests; community members crowdsourced translations the authors do not endorse. The host reports 28,000 GitHub stars at recording and says it answers the question his deep learning students always ask about why to add layers or neurons.

2025-03-12 · open the original ↗

• A deep learning system beat radiologists at predicting breast cancer from mammograms. Godbole, co-first author of the 2020 Nature paper "International evaluation of an AI system for breast cancer screening", says it predicted biopsy-confirmed cancers two or three years ahead, depending on the country's screening guidelines, better than expert radiologists. It was benchmarked on a large retrospective US and UK dataset of what he recalls as a hundred-and-some thousand patients, and beat every radiologist in a separate reader study.
• The bigger promise is combining AI and humans and widening access. Godbole says humans and AI have different strengths, so their synthesis can be powerful, though he is out of date on that literature. He highlights countries with many patients per radiologist, where AI could substantially improve access to potentially life-saving screening.

2025-03-11 · open the original ↗

• Use AI to make yourself more agentic, not to build agents. Godbole, recently retired from Google, argues that once knowledge is commoditized the differentiator is how wise, pro-social and self-authored a person is. He wants AI to support metacognition and a reciprocal loop: acting more, hitting one's limiting beliefs, and growing. He credits Toronto professor John Vervaeke's work on wisdom and relevance as a major influence.
• AI predicted breast cancer from mammograms better than radiologists. Godbole was co-first author of the 2020 Nature paper showing a deep learning system predicted biopsy-confirmed cancers two to three years ahead, depending on the country, on a US and UK dataset of what he recalls as a hundred-and-some thousand patients, and beat every radiologist in a separate reader study. He sees the bigger promise in combining human and AI strengths and in expanding access where patients far outnumber radiologists.
• An LLM must be told which "cinematic universe" it is in. Godbole argues that for an LLM there is no objective fact: pre-training fits the union of all universes in web text, and post-training narrows it to an assistant-and-user universe, so models are statistical models role-playing as human raters role-playing as AI assistants. This first half of his prompting playbook, he says, will outlast its prompt examples.
• Prompt as explicitly as you would brief a stranger. He advises imagining a person off the street with all human knowledge but no knowledge of you, and pasting one's complaints about a failing prompt into the prompt itself. Skills from couples therapy and the book Nonviolent Communication carried over, and he reports an anecdotal inverse correlation between ML expertise and prompting skill.
• Cheaper, better models will unbundle job roles and squeeze undifferentiated products. Extrapolating modestly from current curves, Godbole expects "full-stack employees" such as PMs who arrive with prototypes and user research before engineers are involved. Because copying an undifferentiated product will be cheap, he speculates software could come to resemble the music industry's winner-take-most economics, with discontinuous effects such as small teams not hiring PMs.

2025-03-10 · open the original ↗

• DeepSeek erased the frontier labs' moat, not merely their benchmark lead. Burkov calls it probably the most important event for language models since ChatGPT: a state-of-the-art model trained for what he puts at 5 to 10 million rather than the 20 to 50 million previously assumed, with a published step-by-step recipe that several independent R1-Zero and R1 implementations have already reproduced.
• It is also far cheaper to run. Burkov compares OpenAI's o1 at $60 per million output tokens with DeepSeek's $2, roughly a 30-fold reduction, so that anyone, "your grandmother", can now run state-of-the-art AI.
• Automated verification replaced costly expert-written training data. Burkov argues OpenAI and Anthropic invested heavily, by his estimate billions, in subject-matter experts writing problem-solving dialogues, whereas R1 used binary rewards from checks: does the code compile or produce the right output, does the math answer equal the known value, does the logic puzzle name the right culprit. Hiding a class in a GitHub repo and rewarding a compiling rewrite is one way he says close to a million examples can be generated with no experts.
• Open-weight models are not open source without their data. Burkov's stricter definition, unlike Yann LeCun's, says a model is like a binary: reproducing it requires the training data. The few models released with full datasets are no longer competitive, since today's pre-training uses roughly 15 to 20 trillion tokens versus perhaps four or five trillion in open datasets.