Souk's Notes - Unofficial Summaries
AI Analyst LabThe Hidden Trait That Turns Ideas Into Products Users Love
- Useful products take humility about how much else must go right. The speaker draws this from working in several roles, and says it holds for internal and external users alike. A successful infrastructure product is a team sport.
AI Analyst LabThe PM Reality Check: Scale & Reliability
- The PM role's surprise was the work behind reliable infrastructure. Asked what most surprised them on becoming a PM, the speaker names how much work a scalable and reliable infrastructure takes. They had expected their value to be a former data practitioner's grasp of user workflows, pain points and requirements, which they still call "certainly a superpower of the PM". The clip ends there.
Mr. GlobalDaddy Trump Went To Get A Pack Of Smokes North Carolina, Missouri and Kentucky. Please Share.
- Trump's hurricane rebuilding promise to North Carolina went unkept. After a clip of Trump vowing to rebuild every devastated town and cut through red tape, he cites the state's governor saying North Carolina taxpayers face hundreds of millions of dollars in further cleanup, against over $1 billion for refitting the jet from Qatar and $45 million for a birthday party.
- Silence on tornado victims shows his interest ended with the votes. He notes Trump has not mentioned tornado victims in Missouri and Kentucky at all, and reads this as Trump visiting disaster zones only when campaigning; having won their votes, he argues, Trump is done with them.
Polis Pandit$We Need More Leaders Like Ulysses S. Grant
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnVR Made Women Nauseous — Until Mary Spio Found the Fix
- 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.
GeoHussarIf there was one #general, who could’ve beaten #Napoleon in his prime, it would’ve been #Suvorov
- Suvorov is the one general who could have beaten Napoleon. He cites Alexander Suvorov's record of 60 major battles without a loss, chiefly against the Ottomans and the French in the 1790s, when revolutionary French armies beaten elsewhere repeatedly lost to him in Italy and Switzerland.
- Suvorov's edge was a trait Napoleon lacked: not overrating himself. He lists lightning, unconventional tactics, strict self-discipline imposed on his men, and troop loyalty won by refusing privileges and sleeping where his soldiers slept. Napoleon shared those, he argues, but overestimated himself and underestimated enemies, citing Egypt, Spain, Waterloo and Russia.
- Suvorov admired Napoleon; Napoleon was dismissive in return. He quotes a Suvorov letter saying Napoleon "carries his council of war in his head" while rivals answer to cabinet scribblers. Napoleon, by contrast, said Suvorov had the soul of a great general but not the genius. The two never met in battle.
AI Analyst LabStop Overcomplicating Your ChatGPT Prompts — Do THIS Instead!
- Simple prompts work better; convoluted ones get convoluted answers. The speaker compares prompting to explaining a task to a person. Define the expected inputs and outputs and give step-by-step instructions, and the LLM does better.
Dr. Jon Krohn890: The “State of AI” Report 2025 — with Jon Krohn (@JonKrohnLearns)
- 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 $9 billion.
GeoHussarHow did Vietnam defeat their invaders so quickly in 1979? #vietnam #hanoi #cambodia #vietnamwar
- Decades of continuous war explain Vietnam's one-month defeat of China. He says Vietnam beat the 1979 Chinese invasion in a month because it had fought the Japanese from 1940 to 1945, the French from 1945 to 1955 and the Americans from 1955 to 1975: two generations of war in which nearly everyone was likely involved.
Crosstalk SolutionsZimaBoard 2 - A Homelab in the Palm of Your Hand!
- The ZimaBoard 2 upgrades the original with faster CPU and networking. It moves to an Intel N150 CPU, dual 2.5 Gbps ports (from dual 1 Gbps) and an all-metal case for heat dissipation; it comes in an 8 GB RAM / 32 GB eMMC version and a 16 GB / 64 GB version, with two USB 3.1 ports, Mini DisplayPort up to 4K at 60 Hz, and two powered SATA 3.0 ports for a mini NAS.
- It ships with ZimaOS but runs any x86 operating system. ZimaOS, IceWhale's CasaOS-based personal cloud OS, builds in remote access and RAID; the host lists Windows, Ubuntu, CasaOS, Proxmox VE, pfSense, OpenWrt and Unraid as alternatives.
- The host had Plex, AdGuard Home and Pi-hole running on it. Plex on ZimaOS with an external SSD needed only a media location added to the container; AdGuard Home and Pi-hole were set up separately because both use DNS. The board was on Kickstarter at recording, scheduled to ship in August 2025; the video makes no sponsorship disclosure.
Stock Tok$The Next Palantir...
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnSolving Global Talent Shortages with VR (feat. Mary Spio)
- 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.
GeoHussarChina’s current Navy almost double the size of the US Navy??? 😱 Let’s clear some things up #navy
- China's larger ship count overstates its navy against America's. China has 754 naval assets to the US's 440, but he argues tonnage is the real measure: the US displaces 4.1 million tons against China's 2.8 million, so the US remains clearly first though China is not far behind.
- China's fleet is built for patrol, America's for power projection. China leads mainly in frigates, corvettes and patrol vessels, which he reads as a navy meant to police home waters, while the US Navy is designed to deliver overwhelming force abroad.
AI Analyst Lab7 Steps to Building Production GenAI Apps | Ep28
synopsis
An episode of the Data Neighbor podcast in which Suri, a machine learning tech lead at Adobe, walks through a GenAI product from vague idea to production, using a worked case of a retrieval chatbot. Suri's conclusion is that getting a first answer out of the system is only the feature pipeline. Productionizing really begins there and is dominated by evaluation, along with load provisioning, feedback and the question of whether an LLM was needed at all.
From idea to prototype. Requirements start vague and differ from person to person, so the first task is defining the data. Inputs are usually known; outputs often are not. For the first prototype Suri recommends existing APIs, such as Azure OpenAI or Claude, and a minimal front end in Streamlit or a Jupyter notebook that the team and product managers can try. The prototype should end with a view of what works, what doesn't, what a good output looks like and what success means. A host notes that vibe coding reaches the same starting point but still has to go through the same process to reach production.
The worked case. A host proposes a chatbot that tells staff which internal data table to use, built from table documentation. Suri lays out the RAG pipeline: gather the tables and documents into a database, preprocess them to remove noise, then chunk them. Suri recommends 250–500 tokens as a starting chunk size, tuned through end-to-end experiments. The car-brochure example shows how an oversized chunk lets the model mix up makes and years. Embeddings are mostly used off the shelf, with options such as Cohere's multilingual model or OpenAI's large embedding for long chunks. Vector stores are a choice among Weaviate, OpenSearch and Pinecone.
Query and prompt. User queries arrive incomplete and spread over several turns, so they are reformulated and, when loaded, decomposed. Suri does this with an LLM, and the hosts add that domain terminology may need subject-matter experts. Retrieval can add a reranker where needed. The system prompt should be simple: Suri says 10 critical instructions outperform 100. Step-by-step reasoning builds trust in sensitive domains, and the prompt should guard against leaking itself. The hosts' aside is that clear communication matters as much with LLMs as with people.
Evaluation. Suri calls this where "the movie starts". Human review of perhaps 100–200 cases comes first and recurs after major changes, then automated metrics for minor ones. Accuracy is joined by use-case metrics such as detail, tone and conciseness. Gold answers must include every valid variant, such as two correct SQL queries. LLM-as-judge is now standard: use a stronger model from a different family, at least two where possible, scoring binary or on a Likert scale, while remembering that families can share the same wrong answer. Embeddings are rarely evaluated on their own. Reformulation and retrieval can be, with RAGAS context precision and recall useful for debugging, though teams mostly check the final output. Defining "good" is the hard part, and Suri says it leads into personalization.
What else production needs. Suri warns against using LLMs where classical models suffice, such as classification. The remaining pieces are provisioning for expected input and output tokens, A/B testing, and capturing mostly implicit feedback in a privacy-preserving way with user consent. Cost, latency and fairness are left for a promised part two.
- Prototype on existing APIs and know what success looks like. Suri, a machine learning tech lead at Adobe, says to start from off-the-shelf models (Azure OpenAI, Claude) and to put a very simple front end, such as Streamlit or a Jupyter notebook, in front of the team and product managers for fast feedback. By the end of the prototype stage you should know what works, what doesn't, and what a good output is. Code quality and tests come later, when the project moves to production.
- Chunk size is a trade-off you settle by end-to-end experiment. Suri suggests starting at 250–500 tokens a chunk: 50 is too short, and 5,000 suits only long-RAG techniques. Chunks that are too small lose context, and chunks that are too large blur topics. Suri's example is a single chunk covering Honda, Toyota and Mercedes brochures for 2023–25, which can return Toyota's 2025 price for a question about Honda. Embedding models are mostly used off the shelf, with fine-tuning only for very specific domains.
- Reformulate the query and keep the system prompt short. User queries are lazy, error-ridden and spread over several turns, so Suri reformulates them with an LLM and splits loaded queries into sequential parts. Suri says a prompt with the 10 most critical instructions follows instructions better than one with 100. It can ask for step-by-step reasoning to make answers explainable and add guards against prompt leakage. A host adds that subject-matter experts are needed when users phrase things in domain terms.
- Productionizing is mostly evaluation, not feature work. Suri starts with human evaluation, perhaps 100–200 samples, limited by human time, and then automates it, limited by money. The judge should be a more powerful LLM from a different family, ideally two or more, scoring binary matches or on a Likert scale against gold answers. Suri warns that different model families can give the same wrong answer. RAGAS-style context precision and recall help debug retrieval, but most teams measure only the final output.
- Don't use an LLM where a simpler model will do. Suri calls LLM classification "completely unnecessary" and still prefers more deterministic, explainable classical models. Production also needs provisioning for expected token load, A/B testing, and feedback tracking. Most feedback is implicit ("make this more concise"), and Suri says it should be tracked in a privacy-preserving way, with user consent.
Crosstalk SolutionsThe Most Powerful UniFi Protect Feature You're Not Using!
- Protect's Alarm Manager can trigger on people, objects and events. Triggers include Face ID (known, unknown or specific faces), license plates, objects (person, vehicle, package, animal, or free-text AI Key prompts such as "FedEx truck" at 70% match confidence), line crossing, loitering, NFC or fingerprint scans, sound, UP Sense sensors and system events; stacked triggers like line-crossing-without-a-hard-hat need the AI Key.
- Alarm actions go beyond notifications, including repositioning a PTZ camera. Actions cover app notifications, webhooks, UniFi Access door unlocks, sounds on the smart chime or upcoming horn speaker, lights, and PTZ moves to a preset that return home after a set time. The host found the feature at the bottom of Protect's menu on Protect 5.3.45 and UniFi OS 4.2.10, the latest at recording.
- Slack alerts use a built-in "Slack post" webhook. The host, on Slack Pro at $87 per user per year and unsure whether the free tier works, creates a channel, builds a from-scratch Slack app with incoming webhooks on, and pastes its URL into Protect. He warns the webhook URL is public-facing and should be treated like a password, and suggests "ignore repeated action" at about 10 minutes to avoid spam.
- Custom webhooks to other platforms are not yet dependable. Pushing alarms to Discord did not work natively, and a Make.com relay received Protect's data but failed to post, seemingly on permissions, though the firm's other Make.com and Zapier automations to Discord work fine.
- Protect has no way to test-fire an alarm. Because a custom webhook only learns its payload when the alarm fires, setting up something like an animal or FedEx-truck alarm means waiting for the real event. The host says he asked Ubiquiti for a test-trigger button and hopes it arrives in a future version.
Stock Tok$Mid-Week Movers: Ford Stock...or Lucid? 👀
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnAI Music and Virtual Concerts: How Entertainment Has Evolved (with Mary Spio)
- 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.
GeoHussarWhy are #Russian troops building up at the #finnish border? Is Russia planning an invasion?
- Russia's Finnish border buildup is not an invasion prelude. He reports troops, tents and fighter jets along the roughly 1,000-mile border, smaller than the pre-February 2022 Ukraine buildup. He does not expect an attack because Finland is in NATO, which he argues is the only reason Russia struck Ukraine: one soldier on Finnish soil would bring in all 32 members.
- The Winter War shows Finland can bleed Russia badly. He cites Russian losses of 300,000–400,000 against 70,000 Finnish in the World War II invasion, which Russia won but was humiliated by; he says the Finns surrendered only because they ran out of ammunition.
Dr. Jon Krohn889: AI-Powered Virtual Reality: The Future of Education and Entertainment — with Mary Spio
synopsis
Mary Spio, CEO and CTO of the VR platform CEEK, argues that immersive media combined with AI and blockchain can train people at a scale physical classrooms and aircraft cannot, and pays creators better than advertising does. She ends up holding that as AI floods the world with plausible content, the scarce things become verified, trusted voices and the human work of curating and validating them, which she says CEEK is built to supply.
What CEEK is. Spio describes a social platform where every post is stored as a blockchain asset the creator owns, viewable on Apple Vision Pro, Oculus or a plain screen. The host lists Lady Gaga, Bon Jovi and Dwyane Wade as users and Universal Music, Meta, Apple and Microsoft as backers. AI music, she says, is the marketplace's most purchased item, sold as streaming, sale or licensing rights to labels and film supervisors; solo AI creators own all the rights, which suits smart contracts. Origins. Born in New York and raised in Ghana, Spio watched a moon documentary during a military coup. She returned to the US alone at 16, became a US Air Force satellite technician, studied electrical engineering at Syracuse and Georgia Tech, and worked on SETI. Trying an Oculus at Facebook put her "on the moon"; an Adele concert where, in her telling, 10 million people chased 100,000 tickets suggested remote attendance. The focus has since moved to education.
Paid content over ads. Spio argues ad-based platforms work for maybe 3–5% of creators. A million YouTube views of eVTOL training might earn $3,000, against a thousand people paying 5,000 each on CEEK; BTS made 90 million in three days from a paid concert elsewhere.
Training at scale. Against a host-cited US shortage of 400,000 teachers and a 4.6 million figure Spio recalls, she says one instructor can teach at scale. Putting 100,000 eVTOL pilots in service means training a million; CEEK works with a leading electric aircraft maker whose cost she puts at about 300 an hour against 4,000 in helicopter fuel. Because the brain cannot yet distinguish VR from reality, trainees build real experience: CPR for new mothers, a Baptist Health nursing residency for staff who leave out of low confidence, and intubation practice that some hospitals, she says, now do on paid low-income or elderly people for around $50.
Headsets built for men. Women vomiting at demos led Spio to interpupillary distance: headsets started at men's lower limit, at or past women's upper one. Men, she says, judge depth by size and women by detail. She built a headset to fix both and uses predictive rendering.
Risk and verification. The host raises personalised VR deepfakes shifting political views. Spio cites Musk on certain mass unemployment and an IMF figure, as she recalls it, of 40–60% of jobs affected, and says CEEK is building blockchain validators to check content accuracy, not 100% but better than nothing.
Competing on value. CEEK won Universal Music over better-funded rivals by understanding entitlements and letting it drag and drop log files to port its library to every headset. Spio's philosophy: live from imagination, start before it's perfect, and go where you're valued, as she did when Boeing bought her patents.
- 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.
GeoHussar#China is not going to invade #Taiwan anytime soon.
- If Russia can't take Ukraine, China can't take Taiwan. He argues Russia shares a land border and Soviet-era roads and railways with Ukraine yet holds under 25% of it after three years. China has no such link: it faces the dangerous Taiwan Strait and an amphibious assault, and, citing another creator, Taiwan has stockpiled heavy weapons for 50 years.
- Sanctions on food and energy imports deter a Chinese invasion. As the world's largest food and energy importer, China would likely face strict sanctions on both after invading Taiwan, which he expects would badly destabilise its economy, on top of the diplomatic costs.
Polis Pandit$'Why I Write': Finding Purpose in One of George Orwell's Best Essays
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnQuantum Computing Will Change Everything. How Can You Prepare? (Dell's Chief AI Officer John Roese)
- 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.
Mr. GlobalThe Unshakable Power of Truth PLEASE SHARE!
- Lack of truth is the root of the country's division. He argues the division was built from small lies that grew, which people accept because they offer temporary comfort while chaining them to fear; only a society living in truth can adapt, and justice cannot be pursued through misinformation.
- Speaking truth takes courage; silencing it takes only cowardice. He says every great movement began with someone willing to speak the truth, which is why he stresses it so often on social media.
Dr. Jon KrohnThe New Jobs That AI Will Create (with Dell's Chief AI Officer John Roese)
- 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.
GeoHussarThe Moodys downgrade in American credit is a major loss. #usa #economy #finance #learn #fyp
- Moody's downgrade will raise US borrowing costs at taxpayers' expense. Moody's, the last of the three major agencies holding a top rating, cut it over rising debt and interest costs. He expects existing Treasuries to lose value ($1,000 falling to perhaps $995) and Washington to offer higher yields, with Japan, possibly China and large institutions shorting the debt and buying it back cheaper.
GeoHussarThe Moodys downgrade in American credit is a major loss.
- The downgrade will force the US to pay more to borrow. Moody's cut the US rating over rising debt and interest costs, so all three major agencies have now downgraded US debt. He expects existing Treasuries to lose value (e.g. $1,000 falling to about $995) and the government to offer higher yields to keep selling debt, with Japan, possibly China and large institutions shorting it and then buying in at the higher yield.
- The cost lands on American taxpayers, not on foreign creditors. He notes most US debt is held domestically: the Federal Reserve, mutual funds, savings-bond holders and intergovernmental debt. On debt of about $36 trillion, annual interest is a little under $900 billion. Taxes and new borrowing pay for it, and he expects that bill to keep growing.
- Borrowing from the wealthy instead of taxing them is a loophole. In his reading, the government borrows through Treasuries what it should have collected in taxes from multi-millionaires and billionaires, then pays them interest and repays them in full. His analogy: a $75,000 earner taxed $25,000 who gets half of it back plus interest. Ordinary taxpayers' money funds that interest.
- Tax cuts, trade barriers and OPEC output will make it worse. He argues new tax cuts for the very wealthy mean more and costlier borrowing. He expects the trade barriers that come with the new trade deals to raise unemployment. He says US energy is "already starting to bleed" from OPEC and OPEC+ production increases, and predicts the effects will show in the coming months.
- Only a postwar-style boom can fix it, and it is unlikely. He rules out DOGE and budget cuts, and says raising taxes back to 2016 levels would help. Debt-to-GDP was 120% in World War II against about 125% now. The postwar fix combined infrastructure investment in the hundreds of billions, a doubled minimum wage in 1950 and defense spending near 10% of GDP, against 3% today. He sees Biden-era infrastructure and higher defense spending as possible, but a doubled minimum wage as improbable with a conservative president and Congress and weak unions. Without one, he predicts an even more unequal America: a shrinking middle class and the top 1–5% holding most of the wealth.
Mr. GlobalMAGA Now Blaming Oil Companies For Gas Prices
- Blaming presidents or oil companies for pump prices is wrong both ways. He notes Republicans now blame oil companies, as Democrats did a year earlier, and says gasoline is roughly where it should be given oil and summer seasonality: the savings show in paying about the same as winter rather than the usual ~50 cents a gallon more, with crude down about $10 a barrel since Trump took office.
- Unchecked exports keep US gasoline prices sticky. He argues laws passed by Congress let companies export as much as they like to chase the highest price, drawing down domestic inventories, so crashing crude hurts producers more than it helps drivers.
- Cutting gasoline exports would beat adding refining capacity. He calls both this administration and the last incompetent at lowering prices for relying on "1972" fixes like pushing down oil, says extra refining capacity would just feed more exports, and implies without naming it that one party enabled unchecked exports.
Dr. Jon KrohnWhy the Future of Work Will Be Done by Agent Teams (with Dell's Chief AI Officer John Roese)
- 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.
GeoHussarChina got caught putting kill switches in the worlds, solar energy power grid
- China's renewables push is about energy independence, not environmentalism. Replying to a viewer, he agrees China leads and may get closer to carbon neutrality by 2060 than the US, which gets about 23% of its electricity from renewables. He reads the motive as escaping dependence on OPEC+ oil and gas, and on dirty coal. The captions never reach the kill-switch claim in the title.
- America's 1970s solar push followed the same logic after the embargo. He compares it to the US response to the Saudi-led OPEC oil embargo, when Jimmy Carter created the Department of Energy in 1977 and the US led the world in solar panels.
Scott HanselmanIs AI what Robotics needs? with Jasmine Lawrence Campbell
synopsis
Jasmine Lawrence Campbell is a technical product manager at a robotic-massage company and previously worked at Microsoft and Nvidia. She ends on augmentation over automation. The useful robots are narrow ones that take the dull, dirty or dangerous parts of a job while a human stays in control. She is wary of general-purpose humanoids and of the idea that plugging in an LLM solves robotics.
Why the home is hard. Hanselman asks why his phone is powerful while his robot vacuum is "dumb". She answers that factories are static and regulated and homes are not. Home devices have limited on-board compute, and their sensors are often used for a single purpose. SLAM gives a pirate-map outline that is rebuilt daily and does not distinguish movable objects from walls.
What intelligence actually involves. She argues people underestimate their own proprioception, and so call robots stupid for failing at what humans do unconsciously. Missing senses include touch texture, temperature and smell. Hanselman's story of his vacuum refusing a door that had once been shut leads her to human-in-the-loop correction, which she compares to coaching a toddler.
Designing around robots. Instead of asking a robot to carry grocery bags the human way, she would change the task, putting the groceries in a cart a robot can lift. She cites curb cuts and other accessibility adaptations as precedent.
Humanoids, LLMs and transfer. She is skeptical of humanoid demos that do not say whether they were sped up or teleoperated. She prefers narrow domains plus transfer learning, such as moving from vacuuming to painting a wall. LLMs suit dialogue but not turning words into actions and replanning. Specialist models per skill are more auditable but still need a central decision layer.
Augmentation. Automation needs task analysis few companies do. She sees robotic massage as a collaborator handling warm-up work for therapists who see 10-plus clients a day, and as a way to serve patients who need frequent sessions. She does not see it replacing the human relationship. Her resources for newcomers are Adafruit, KiwiCo, and Udacity and Coursera, where she teaches an AI product management course.
- Home robots lag because homes change, not because hardware falls short. Lawrence Campbell, a technical product manager at a robotic-massage company, contrasted constrained, regulated factories with homes where shoes and rugs move daily. Robot vacuums are limited by on-device compute and by sensors dedicated to one job, such as lidar used only to detect ledges. Their SLAM maps are rebuilt each day without any sense of which objects move.
- Robots look dumb because we underestimate our own intelligence. People have no conscious sense of the proprioception they use to navigate. She said the hardware may be ready while the software is "way farther away than we think", that touch and smell remain barely digitized, and that robots need human-in-the-loop coaching the way toddlers do.
- Change the environment to suit robots rather than demand human dexterity. Her example was a cart a robot could lift whole, rather than expecting it to thread grippers through grocery-bag loops. She tied this to accessibility measures such as curb cuts, on what she called roughly the 35th anniversary of the US ADA laws.
- Narrow robots plus transfer learning beat humanoid demos. She distrusts humanoid videos that do not disclose speed-ups, test conditions or teleoperation. She favors mastering one domain and transferring the skill, for instance from pushing a vacuum to painting a wall. LLMs help with dialogue, but she argued they do not close the gap from words to actions and constant replanning. A set of specialist models is easier to audit but still needs a central decision-maker.
- Augmentation beats automation, including in massage. Full automation needs a depth of task analysis she said most companies will not do. She sees her company's robot handling warm-up work while a therapist does the detailed work, extending therapists' careers and serving people, such as lymphedema patients, who need massage several times a day.
Polis Pandit$How To Use the Socratic Method In Your Own Life
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon Krohn888: Teams of Agents: The Next Frontier in AI Collaboration — with Mike Pell
synopsis
Mike Pell, director of the Microsoft Garage, speaking live at the AI and Creativity Summit in Brooklyn, argues that AI is an exoskeleton that will soon act as a proactive, trusted co-worker, that the hard skills of most jobs will be automated, and that people's value will lie in communication and collaboration. He ends as an optimist: "we're getting to the good part".
Innovation without permission. Pell says Microsoft, since Nadella became CEO, chose to make curiosity cultural rather than grant 20% time: people are told not to ask permission, and to return with data.
From passive to proactive. Pell calls AI an Iron Man suit and agents the packaging that makes it feel like an adviser. He predicts assistants will soon interject unprompted, and warns users not to stop verifying once trust is earned. GitHub Copilot, he says, nearly died over early syntax errors; today Garage clients see three-day research tasks done in 30 seconds.
Teams of agents. Chaining tasks is the current state of the art, Pell says; agents completing ambiguous tasks together, in "swarms", is not close. The host warns that a 1% error rate compounds across 10 agents; Pell counters that review agents and orchestration could reduce it.
Soft skills and ubiquity. Pell says graduates lack debate and critical-thinking skills, and that AI should make people better teammates since engineering or medical knowledge will come from the system. He argues AI is already everywhere and will soon stop being called AI.
Audience questions. Asked how to keep up, the two suggest crowdsourced sentiment and the LMSYS arena. Asked about lost jobs, Pell cites Microsoft's book The Future Computed and history's precedents, urging people to focus on what they will do about it.
- 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.
GeoHussarRussia will pay for insulting the Turkish President by not participating in the peace talks. #russia
- Putin's no-show at peace talks insulted Erdogan, and Russia will pay. He reports Zelensky attended while Putin, who had proposed restarting direct talks in that city, sent a low-level delegation, and the US president then also stayed away. He quotes Zelensky calling it personal disrespect to Erdogan, and predicts Russia will suffer because Turkey is the region's main military and diplomatic power broker.
Crosstalk SolutionsFortinet + UniFi: Seamless Network Integration Guide
- UniFi switches and APs work behind a FortiGate gateway. The demo uses a FortiGate 70F, a UniFi Flex switch and a U7 Pro, continuing the channel's FortiGate small-business setup video, and the host says the approach scales to as many UniFi switches and APs as needed while keeping the FortiGate's UTM features.
- A FortiGate DNS entry lets new devices find a cloud controller. This is optional if a local controller sits on the same subnet, and set-inform over SSH also works. Otherwise, enable DNS Database under Feature Visibility, have DHCP hand out the FortiGate's .1 address as DNS, create a recursive DNS service on the internal interface, and add a "unifi" zone whose "@" entry points to the cloud controller's IP.
- Already-running devices must reboot to pick up the new DNS. In the demo the U7 Pro appeared for adoption but the switch did not until both devices were rebooted and pulled fresh DHCP leases.
- In UniFi, only VLAN IDs matter; the default subnet can stay wrong. The controller's default network shows the wrong subnet and can't be changed in the current version, which the host says is harmless with no UniFi router. Employee (VLAN 20) and guest (VLAN 30) are added as virtual networks with "third-party gateway" as router, then SSIDs are mapped to them.
- Switch ports carry the VLANs the FortiGate defines. The FortiGate uplink and AP ports keep the default native VLAN and allow all tagged VLANs, since the AP needs both to serve its SSIDs; an access port, here port 2, is simply set to the guest network.