Souk's Notes - Unofficial Summaries

2025-12-22 · open the original ↗

• Ives says Nvidia chip demand outstrips supply 12 to one, based on a recent three-week Asia trip. He argues Nvidia is four to five years ahead of rivals such as Google's TPU, Huawei and AMD.
• He argues denying Nvidia access to the China market would only strengthen Chinese domestic competition. This is his own read on why continued China access matters for Nvidia's position.
• Ives frames his AI bet as a bet on Jensen Huang specifically, calling him the "godfather of AI." He compares this to past leadership-driven calls he got right, including the iPhone being dismissed as "a one-year AT&T smartphone" in 2008, skepticism of Satya Nadella's Microsoft cloud pivot in 2014, and doubts about Elon Musk's electric vehicles.

2025-12-21 · open the original ↗

• Data is now a means of production alongside capital and labor. Pentland cites Xi Jinping as saying it is "capital, labor, and data", and that whoever controls data can manipulate the other two, which he says is where the hyperscalers' power comes from; he adds that Xi "apparently did get it from me." Individuals, he argues, are at the mercy of firms that know much about them while they don't know what those firms know.
• Personal agents can get results without surrendering personal data. Pentland says an agent can query what is out there and ask specific questions without disclosing your age, sex or income, which makes it a chance to "take back your data."
• Data cooperatives can counter corporate concentration, as banks and unions did. He points to mid-1800s agricultural cooperatives that broke the Eastern banks' monopoly when banking was confined to New York, Boston and Philadelphia, later built much of America's electric grid, and evolved into today's credit unions, and to labor unions of the early 1900s. Through the Consumer Reports loyal agents project, pooled questions and answers would let members spot scams, find deals and compete with firms like Walmart.

2025-12-21 · open the original ↗

• Britain's eighteenth-century wars were mostly strategic failures. The War of the Austrian Succession (1740–1748), fought alongside Austria, German states, the Dutch Republic and Russia, ended in a draw he calls a strategic failure because it cemented French power and territory. He says Britain lost the American Revolution (1775–1783) to Americans backed by France, Spain and others, and lost twice in the French Revolutionary Wars (1792–1802).
• Britain still emerged dominant, and alliances were why. Despite losses anyone would have read as decline, he says Britain came out of Napoleon's defeat and the Congress of Vienna in 1815 as the uncontested power for about 100 years. He names alliances as one of its advantages, framing Britain as the precedent for the United States; the captions end there.

2025-12-21 · open the original ↗

• Winning major wars through allies shows strength, not decline. He accepts that the US has not won a major war alone since the Spanish-American and Philippine-American Wars, setting aside short interventions such as the Dominican Republic (1965), Grenada (1980s) and Panama (1989). He rejects reading this as weakness, arguing alliances are what history shows great powers use.
• His examples are both world wars and Korea. In both world wars the US fought alongside Britain and France, and in the second also alongside the Soviet Union. Korea (1950–1953) was a UN coalition. He calls Vietnam one of America's greatest failures, but the captions end before he develops the point.

2025-12-20 · open the original ↗

• Emotional suppression is taught, and it crosses cultures. Atanga, a physician in internal medicine and pediatrics for 30 years, almost 20 of them with the VA, says patients from tech arrive with insomnia, high blood pressure or stress, and underneath she finds suppression learned from boyhood ("men don't cry"). She has seen it in Africa, where she was born, in Europe, where she went to medical school, and in the US. She says women who move into traditionally male spaces such as tech are expected to suppress as well, and she argues that society empowered women and children but "forgot the men".
• Tech promotes people into management without the emotional skills. Atanga says people are promoted for their core skill, such as coding, and leadership training then targets measurable, box-checking goals rather than handling their own and others' emotions. People quit over a toxic environment or a poor manager, not over the job, and she ties burnout and resignations to this. Giard adds the company's side: turnover means expensive retraining and no continuity.
• Suppressed emotion builds pressure that surfaces later as illness or divorce. She points to leaders who "drop dead at work" and couples who divorce after 30 years of marriage once one of them retires. People get checkups for back or knee pain, she says, but nothing checks emotional health until a crisis. Her analogy is a balloon: suppressing does not push the feeling down, it inflates the balloon until it pops.
• Feelings drive thinking, and thinking drives behaviour. Using the "feel, think, behave" arrow from her talk, Atanga says people feel justified in acting on a feeling about a colleague, so the fix is to ask where that feeling came from. If it is justified, as with a person who is always aggressive, she says to set boundaries that still let the team be productive. She says tech teams are now more diverse and tech is everywhere, citing a keynote from Little Caesars at Michigan DevFest, so leaders end up working with people unlike themselves.
• The fix is a no-fault culture change, brought in from outside. Atanga says the culture created the problem, so no one should be blamed, and workplaces should make healthy expression and time to heal acceptable, as a sports team rests an injured player. She says bereavement leave averages about three days, and her residency gave three, or five if travel was over 5,000 miles. Another employer gave none, so the time came out of sick leave. She says employees avoid in-house counsellors because using one identifies them, and divorce can make colleagues read a person's performance as slipping. Employers therefore bring in outside speakers like her to give everyone tools, with personal help available beyond that.

2025-12-20 · open the original ↗

• Current AI agents can't be trusted to serve their user. Pentland says agents raise the question of whether they serve Amazon or you, misbehave on edge cases, sometimes do illegal things, have weak guardrails, and "get lost" in networks of agents, forgetting their task. The open-source loyal agents project with Stanford Law School aims at agents that fulfil the user's intent legally and reliably.
• AI buddies that direct attention beat AI that decides. He describes a company, mentioned by a CTO on a panel he sat on, that gave every employee an on-premises AI buddy which has read the manuals and newsletters and tracks what other departments do. He says such a buddy takes a good team a few days to build and changes company culture, especially for remote and distant offices. If the AI is telling you what to do, he says, "you have made a mistake", because agents won't be reliable enough to decide for the next several years.
• Deliberation without big voices undercuts polarization. Pentland's open-source deliberation.io works like X without dominant accounts: it visualizes what everyone thinks and adds an AI mediator that summarizes themes in a Socratic style. He argues most Americans agree even on abortion and gun control, and polarization comes from picturing "the crazy guy with five million followers." He reports that in experiments people liked the AI summaries twice as much as human ones, even with a cheap model.
• Residents asked for AI to defend them against the city. In a deliberation exercise with residents of Washington, DC, on AI in city government, Pentland says what busy residents wanted, unexpected by the experts, was personal AI to navigate regulations, fill out forms and find opportunities.

2025-12-20 · open the original ↗

• Measured against US invasions, Russia is failing in Ukraine. The US invaded Afghanistan on 7 October 2001 and Kabul fell on 13 November; it invaded Iraq on 20 March 2003 and Baghdad fell on 9 April, despite fighting halfway around the world. Russia, bordering Ukraine, holds about 20% of the land after three years without the capital and, by his figure, has lost 1.1 million men.
• The US failed at nation-building, not at beating armies. He concedes the US did not build lasting democracies in Afghanistan or Iraq, but argues it reigns supreme at defeating enemy armies, which Russia keeps failing to do. He credits the comparison to Adam Kinzinger.

2025-12-19 · open the original ↗

• US oil output figures may be overstated, threatening higher fuel prices. His worry rests on the latest Dallas Fed energy survey, which he says found the official US production data cannot be correct. He frames the risk as a collapse in the oil industry feeding through to gasoline prices and energy bills.
• Record output makes no sense at $56 oil. He cites US production of 13.84 million barrels a day, 3 million more than Saudi Arabia and 4 million more than Russia. Historically the US raises output only when prices rise, and he asks why new wells would be drilled at today's price. The captions cut off mid-argument.

2025-12-19 · open the original ↗

• This episode is a holiday greeting, with no substantive content. The host and the show's staff, including its founder, podcast manager, partnerships manager, media editor, writer, researcher, shorts editor and thumbnail designer, each introduce themselves and wish listeners happy holidays, with a promise of more data science, machine learning and AI content in the new year.

2025-12-19 · open the original ↗

• Train attributes 2025 underperformance to a structural growth bias that missed a value rally. He names London Stock Exchange Group and Diageo as portfolio holdings down year-to-date, whose growth credentials he believes have come into question only temporarily.
• His response to a possible mistake is to hold through difficult periods rather than sell, betting on enduring franchises. He concedes this patience may be "to a fault" but says it's deliberate.

2025-12-19 · open the original ↗

• Career earning power peaks early, then flattens and declines. Responding to a viewer who noted this contradicts the assumption of ever-rising pay, one host says national research shows a distribution curve: workers sit below their peak value in their 20s, peak in their mid-30s, plateau, and decline through their 50s and 60s -- comparable to how sports leagues pay athletes their highest salaries in their 30s despite peak physical performance coming earlier.
• Add wings once a strangle eats too much buying power. For a caller with a $200,000 account and $100,000 committed to trading, the rule of thumb is that a naked strangle should use no more than 5-7% of trading capital -- $5,000 to $7,000 here; once it exceeds that, buy wings wide enough to bring the requirement down to roughly $2,000-$3,000, rather than adding contracts or overthinking it further.
• Bet mostly on yourself, not a trading account, with windfall cash. A profitable trader who won a lawsuit settlement asked whether to invest it in his own company or trade it first; the host says to put the bulk into the business and keep only 10-30% for trading, comparing it to why companies buy back their own stock -- though he notes this mentality mainly applies to founders whose upside is tied to equity, not hired executives on salary.
• Compounding is oversold for traders; basis reduction matters more. Asked about compounding as a wealth-building strategy, one host calls it "a bit of a sales gimmick" for someone trading capital rather than investing passively, saying his own results come from lowering cost basis through selling options and taking asymmetric risk; his co-host pushes back that compounding is real math (the rule of 72) for buy-and-hold investors, just not for traders who need dry powder.

2025-12-19 · open the original ↗

• Lack of investment, not sanctions, sank Venezuela's economy. He dates heavy US sanctions to 2005 and Chávez's nationalization of all oil infrastructure, forcing US companies out, to 2007. GDP held up for a few years, then fell to about $111 billion, roughly 26% of its level a decade earlier.
• Even Russia and China would not rebuild Venezuela's oil sector. He argues the crash persisted because no one would invest in so unstable a country. Russia and China had the capability but, on his reading, did not trust the region's stability.
• Maduro survived the US-backed challenge and turned aggressive over Essequibo. Maduro succeeded Chávez in 2013 and still holds power, despite US-backed Juan Guaidó's 2019 claim that the 2018 election was rigged. In the 2020s Maduro pushed to reclaim Essequibo, which he notes has been part of Guyana for over 130 years.
• A conflict looks imminent, he says. As of December 17, 2025, he says the US has massed its largest naval concentration off Venezuela's coast, and "it looks like some kind of conflict is about to start."

2025-12-19 · open the original ↗

• Venezuela was a close US partner from 1945 to the 1980s. A world oil shortage after World War II made Venezuela's oil valuable, and he says it was seen as the model strong, stable, wealthy Latin American country, able to buy US F-15 and F-16 fighters.
• US collusion with OPEC to crash oil broke Venezuela. He says that in the mid-1980s the US worked with OPEC to drive oil prices down to bring down the Soviet Union, crushing the Venezuelan economy. He argues the crash and slow recovery are what let socialism take root.
• Chávez's failed 1992 revolt preceded his rise to power. He says Hugo Chávez openly admitted what he had done without remorse, that Venezuelans loved him, and that by 1999 he took power; the captions cut off mid-sentence.

2025-12-19 · open the original ↗

• The US let Venezuela down over Essequibo. Venezuela invoked the Monroe Doctrine against British expansion, and by 1897 the US favoured Venezuela's claim. Facing a choice between war with Britain and arbitration, it chose arbitration, and the 1899 Paris tribunal unanimously awarded Essequibo to British Guiana. He says Venezuelans felt betrayed.
• Oil profits kept US–Venezuela ties going despite the betrayal. He says both sides profited as the US built oil infrastructure in oil-rich Venezuela. The relationship intensified under the dictatorship of Juan Vicente Gómez, whom he calls brutal, until 1935.

2025-12-19 · open the original ↗

• Britain rejected Venezuela's Essequibo claim as an empty Spanish line. Venezuela took the Spanish colonial borders, including Essequibo, at independence in 1823, and US–Venezuelan relations were cordial until the 1895 dispute. Britain objected that the line was drawn by a Spain no longer in power and that the region had no Spanish or Venezuelan settlement. The captions end before his third reason.

2025-12-18 · open the original ↗

• Venezuela's 1823 victory marked the Spanish Empire's true death. That year the US became the first country to formally recognise Venezuelan independence and proclaimed the Monroe Doctrine. He presents the remaining colonies' fall as only a matter of time from then on.
• Britain backed the Monroe Doctrine because it served British interests. He argues Britain, though no friend of the US, welcomed the doctrine because it kept Portugal, France and Spain from retaking lost territory. That let Spain's remaining colonies break away without Spanish troops, with Bolivia's bloody 1825 war as the exception.

2025-12-18 · open the original ↗

• Agent runs are turning stakeholders into code collaborators. One speaker, describing their company's own production code bases (Netlify, from context), says product managers, designers and support engineers now open agent runs to prototype a fix, then push it to GitHub for the dev team to integrate. The speaker says PMs, and even CEOs, are taking changes to production, and that extending development to the wider team is part of their vision.

2025-12-18 · open the original ↗

• Venezuela lit the fuse of Latin American independence. He says Venezuela was the first to declare independence from Spain, in 1810, and every other Latin American country followed within the year. Spain, badly weakened by the Napoleonic Wars, tried to make an example of its resource-rich colony. The early revolutionary leaders could not unite until 1813, when Simón Bolívar stepped in, and the captions end there.

2025-12-18 · open the original ↗

• Agents get more attackable as they get more flexible. Song, a UC Berkeley computer science professor and co-director of its Center on Responsible Decentralized Intelligence, says her group's overview paper maps agentic systems along dimensions such as static versus runtime-selected tools, autonomy and control-flow flexibility, and that each step towards flexibility widens the attack surface.
• Agentic AI worsens security on both sides at once. The agent itself is harder to protect, since prompt injection or jailbreaks could make a coding agent delete a database or leak files, and attackers can misuse capable agents against other systems; she points to her group's cyber-security evaluations such as CyberGym. Her stated goal is provable guarantees of security properties before agents take critical actions.
• Nobody understands how the LLMs powering agents work. Song says we hand privileges, and perhaps one day credit card numbers, to systems that hallucinate and can be jailbroken, without knowing when they break. Hanselman framed this as the internet becoming "one giant text box" whose input we are now expected to trust.
• LLMs have jagged intelligence and generalise poorly under complexity. Asked about the stochastic-parrot view, she cites math-olympiad gold medals alongside silly mistakes; her group's controlled benchmarks find generalisation under fine-tuning and reinforcement learning degrades as difficulty and compositional complexity rise. She says nobody knows whether the transformer-plus-RL paradigm is sufficient.
• Agent evaluation needs open, standardised, reproducible benchmarks. Her agentic AI MOOC, started fall 2024 and now in its third edition, has over 32,000 enrolled; its spring competition drew close to 1,000 teams. The current competition targets agent evaluation under a paradigm she calls Agentified Agent Assessment (AAA), with over $1 million in prizes and resources from sponsors including Google DeepMind.

2025-12-18 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2025-12-18 · open the original ↗

• Use neural networks when data outgrows classical machine learning. Guha Biegas, a data analyst programmer at Freudenberg, says forecasting one household's next energy bill from its smart meter is a job for ordinary machine learning. Forecasting a whole area's usage calls for deep learning, which trains in batches in parallel on GPUs, so you can inspect, debug and fine-tune before training finishes. She names multimodal data (image, audio, video, language) as the main use, because RGB data is so costly to process.
• Accuracy that is too high is a warning sign. On her battery work she used available PyTorch models for survival analysis of lithium-ion cells, and says they gave accurate results about 80% of the time on new data. She treats accuracy that is too high as a red flag that the model is overfitting.
• Watch training visually, not just as console metrics. Her Michigan DevFest workshop trained a convolutional network with PyTorch on the Fashion-MNIST dataset: 60,000 grayscale training images in 10 classes such as coat, sandal and ankle boot. She logged the run to TensorBoard. She says she is a visual learner and that accuracy and precision figures in a console do not show a boss how far a problem is from solved. Clicking into TensorBoard's live graph of the network let her "go inside the black box", see when the model learns each class, and compare learning rates and momentum against the loss.
• Check class balance with TensorBoard's projector before the model overfits. She used its principal component analysis view, coloured by label, to see how the 10 classes are spread and whether some are fed far more than others; the variance shown was 54.7. She says this matters more when a dataset has over 100 classes. She has not tried the UMAP or t-SNE views.

2025-12-18 · open the original ↗

• Today's AI could fail society the way Soviet planning did. Asked by the host whether the parallel holds, Pentland says "absolutely", for the same reasons: bad data, failure to account for fast change, and no attention to how people actually work.
• Humans are social foragers who mostly copy others. Pentland says nobody reads manuals; people watch others and ask a question or two. He cites Daniel Kahneman's estimate that probably 95% of what we do comes from the culture around us, which he warns "doesn't mean it's right", with the remaining 5% being what makes you you.
• Systems that treat people as individuals produce echo chambers and loud voices. He argues social media matches people on similar feature vectors and ignores community, and its follower-and-advertising model means only the loudest, most extreme voices get heard. He says his group has shown "very definitely" that this is where polarization comes from, and that removing big voices reveals broad agreement, in corporate decision systems as well.
• Watching other people beats following the equations alone. In experiments with traders handling $10 million a day, Pentland says those who paid attention to other traders reliably did better, because people catch the edge risks and changing conditions that models miss. He reads 2008 the same way: models with unnoticed tail risk nearly took down the system, while those also watching human behavior came through.

2025-12-18 · open the original ↗

• Recording is the easy part; editing and distribution eat the time. The three Data Neighbor co-hosts, who met at Nextdoor, say that even a rough edit or a bad thumbnail takes longer than it should. Shane, who does the thumbnails, says no one warned him about the YouTube work. A co-host says the weekly back-office work of publishing to Spotify, Apple, Substack and LinkedIn takes far longer once it compounds.
• Busy, well-known guests say yes if you are consistent and courteous. A co-host says a prominent statistics educator agreed to appear when the show had nothing to show, and that they spoke to 10 CEOs and co-founders in the month before recording, some from large companies. A co-host says the free advice guests give would otherwise cost thousands.
• The podcast's purpose shifted from helping pivoters to learning AI. The co-hosts started out to help people move into data, but a few episodes in they decided the AI shift was faster and more consequential than was being discussed. They say they ran out of their own knowledge in about three episodes and now use the guests to upskill themselves. Shane's turning point was an episode on building products with Replit, whose guest teaches paid courses on it.
• Three co-hosts was a deliberate hedge against burnout. A co-host says the odds of one person burning out or quitting are high, so two is better than one and three lets them rotate. Shane adds that each episode has an informal lead and a support, so the second host can always fill a lull in the conversation.
• At principal level the job reaches beyond your company. A co-host who is at principal level describes a ladder: a junior is handed work, a senior IC tells the manager what the team needs, staff works across teams, and principal must bring the industry's practice in and push the company's out. A weekly podcast, the co-host says, forces that outside contact. The co-hosts also credit it with more visibility, more inbound recruiter messages and more structured speaking, which one co-host estimates has improved 10x to 20x.

2025-12-18 · open the original ↗

• MSTR now trades like a closed-end fund below its own Bitcoin. One host has argued for years that the company's convertible-debt strategy, issued without a set interest rate or conversion ratio, was "a death trade"; with roughly $50 billion in Bitcoin against a $47 billion market cap, he calls it a reasonable bounce trade for someone bullish on Bitcoin, but warns a continued Bitcoin selloff could force further asset sales, creating "a vicious cycle."
• Credentials alone won't buy trust with older advisory clients. A Purdue finance senior finishing a CFA track and wanting to both analyze stocks and build a client book is told his age, not his degree, is the obstacle -- a 22-year-old advising clients decades older faces "what does this kid know?" skepticism regardless of qualifications, so the priority should be joining a firm that exposes him to multiple paths (analyst, adviser) before committing to one.
• Bad hires should go fast, though the hosts admit they don't. Answering a founder stuck with an underperforming employee, they say a wrong hire costs more than no hire in money, productivity, and the founder's own bandwidth, but confess their own pattern over the years has been to sideline problem employees rather than fire them -- calling it "not a scalable model."
• Nvidia has topped out but is too big to fail. One host, bearish on the stock since a disastrous short years ago when it traded near a split-adjusted $15, argues the last earnings report was "too over the top" and expects a wide trading range rather than a crash; for a long holder questioning the position, his suggestion is to sell calls against it to lower cost basis rather than sell the stock outright.

2025-12-18 · open the original ↗

• AI is removing the gate that made coding a priestly skill. One speaker compares today's developers to pre-printing-press scribes, a select group able to read sacred texts, and predicts billions of people will be invited to do things with computers. The speaker argues developers' importance only grows, but that software development will become part of every job the way writing did, while "you probably can't get a job as a scribe" today.

2025-12-18 · open the original ↗

• Select London-listed world-class companies are underowned and undervalued. Train concedes the UK has disappointed because its benchmark lacks structural growth exposure, and won't say that is about to flip. He would not recommend a UK ETF; the opportunity he sees is in rare, globally significant businesses that offer under-recognized access to wealth created by AI.
• Owners of proprietary data are AI's strategic winners. He borrows LSEG CEO David Schwimmer's framework: compute will commoditize and agents will battle endlessly, while proprietary data holds the advantage, citing LSEG's data deals with Anthropic and OpenAI. Relx's CEO estimates its AI-driven risk division went from 2% of the company's value 20 years ago to over half today. The NASDAQ lesson, he says, is that the starting multiple matters little for capital-light digital businesses with a large addressable market.
• Diageo's enduring brands beat technology that risks obsolescence. He cites Bernard Arnault's view that Dom Pérignon, unlike the iPhone, will still be desired in 50 years, and argues that a 1999 investor choosing Nokia over Diageo would have been wrong. Finsbury bought Diageo at £5 in 2000 and saw £40 by 2022; consumers drink less but better, and Guinness, about 15% of Diageo, keeps growing.
• The 2025 underperformance reflects a value rally, not a broken thesis. Finsbury is down 3% year-to-date against the benchmark's 20% gain, with LSEG and Diageo both down, though since December 2000 it is up 733% against 317%. He is responding by buying more Finsbury shares; turnover has averaged under 5% a year, and the top 10 of about 20 holdings make up 80–85%.
• Optimism is a trained discipline, and pessimism breeds failure. He says he consciously chose about 30 years ago to stay optimistic, having seen caution sink investment careers, and treats setbacks as buying opportunities. His mantra is "be bold, be disciplined, be optimistic"; he urges Finsbury shareholders to back the trust's continuation vote in January.

2025-12-17 · open the original ↗

• ARC Prize defines intelligence as the efficiency of learning new skills, not the ability to execute known ones. Its president explains the foundation built on François Chollet's 2019 definition, contrasting it with benchmarks like MMLU that chase harder "PhD-level" problems; ARC tasks are deliberately kept solvable by ordinary people, testing whether a system can generalize rather than just score high.
• Reasoning models produced a step-change on the ARC-AGI benchmark that pure pretraining never did. GPT-4's base model scored only about 4-5% as of 2024, but o1 and o1-preview jumped to roughly 21% once reasoning was introduced, which the foundation says is why OpenAI, xAI, Google, and Anthropic have all reported ARC-AGI scores for their frontier releases (Grok 4, Gemini 3 Pro/Deep Think, Opus 4.5) over the past 12 months.
• The president warns against mistaking narrow RL-environment scoring for real progress. He argues that building a custom RL environment to ace a specific benchmark doesn't generalize the way humans do without one, calling it "whack-a-mole," and says ARC guards against this with a hidden test set and an emphasis on novel problems.
• ARC-AGI-3, due next year, will be an interactive benchmark with no instructions and will score efficiency, not just accuracy. Test-takers must infer the goal of roughly 150 game-like environments purely from action and feedback, and performance will be normalized against average human action-counts rather than raw success rate. The president maintains that even a hypothetical 100% score would be necessary but not sufficient evidence of AGI, prompting further scrutiny of the system's remaining failure points rather than a declaration.

2025-12-17 · open the original ↗

• AI fails from bad models of society, not bad algorithms. Pentland, whose book is Shared Wisdom, says his decades of measuring how people interact, with wearables, tone of voice and movement, show that systems designed as if people were logical engines make "bad mistakes"; the system to design is AI plus people plus society.
• The Soviet Union's optimization-based planning failed on data and change. He says the first wave of AI, optimal resource allocation through linear constraints, still the most common computation in the world, was used to run the Soviet economy. People lied about the numbers, conditions changed, and rare events broke the models, and he says this is thought to have been a major contributor to its collapse.
• Expert systems optimize for the average and hollowed out communities. Pentland says rule-based systems now run corporate approvals, audits and the tax system: cheaper, but inflexible, and worse the further you are from average. He links the centralization they enabled to the loss of local clinics and local banks and to "bowling alone."
• Machine learning repeats the problem by looking backward. It needs the right data and lags reality, which he says produces social-media echo chambers and decisions "two heartbeats behind" that can kill a company. Systems must be designed around users, buyers and bad actors alike.

2025-12-17 · open the original ↗

• Crypto's high volatility makes a small allocation mathematically justified, not reckless. Answering a self-doubting newbie, one host argues implied volatility in Bitcoin (roughly 50%) and Ripple (roughly 100%) versus the S&P's roughly 17% offers a four-to-five-times asymmetric upside opportunity; the recommendation isn't 10-50% of a portfolio but something like 0.5-2%, held long-only since most retail accounts can't short or hedge crypto anyway.
• Lead with AI now; entrepreneurship value peaks later. A University of Illinois finance junior interested in both AI and eventual entrepreneurship is told fresh AI knowledge is worth the most immediately out of school, while entrepreneurial value peaks in the early-to-mid 30s; the host also recommends taking the standalone SIE licensing exam before graduating as a cheap, low-effort signal of initiative to employers.
• Advisor pushback is about fees, not your risk. Asked how a retiree can convince an investment adviser to let them self-direct part of their money, one host explains resistance usually evaporates once a client threatens to move the account, since advisers are paid on assets under management -- often 55 to 100+ basis points -- rather than performance; he recounts asking an adviser how returns compared to the S&P and getting "we don't compare ourselves that way" as the tell that it was time to leave.
• High IVR argues for defending, not abandoning, a losing silver trade. One host, short in-the-money silver calls from the low-to-mid 50s through the low 60s while the metal keeps rallying, says implied volatility rank around 85-87% argues for holding rather than closing at a loss -- staying small on delta and spreading with far out-of-the-money short puts -- and only cutting part of the position if it becomes genuinely unmanageable.

2025-12-17 · open the original ↗

• A commit graph shows exactly when someone moved into management. One speaker made a graph of their GitHub commits and could see the point their activity died off after becoming a manager; the other recognised the same drop in their own graph, and the explanation offered was "I was in meetings".

2025-12-16 · open the original ↗

• n8n automatically restarts his UniFi OS Server when its CPU locks up. The server hangs every four or five days. Uptime Robot detects the outage and sends a webhook, which is rate-limited to one run per 10 minutes. n8n then SSHes in with a private key, as a service user allowed sudo only for the UOS commands, to stop, start and status-check the services, and logs each run to Google Sheets.
• A RAG workflow suggests answers to new support tickets. At Rogue Support, technicians paste a ticket into a Cloudflare Pages form. A webhook passes it to an AI agent (ChatGPT) backed by a vector database of hundreds of past tickets, which returns questions to ask, a likely fix, a time estimate and gotchas in about 15–20 seconds.
• An LLM résumé scorer took 2 hours to build. It drops international applicants, then scores each candidate out of 100 on customer service, résumé, stated knowledge and availability. It writes Google Doc summaries for those scoring 81–100; in the latest hiring round it cut over 50 résumés to the top 15.
• His most complex workflow filters RSS news into Discord. Every 6 hours a parent workflow loops through feeds listed in a Google Sheet. A child workflow has Anthropic's model score each new article 1–5 against his topics (cybersecurity, networking, VoIP, AI) and posts those scoring 4–5 to Discord with a category, link and bullet summary.
• n8n is free, self-hosted in Docker, and LAN-only by default. His runs in Container Manager on a Synology NAS, reached over HTTP on port 5678. For internet access he suggests a Caddy plus Let's Encrypt proxy, a DigitalOcean one-click install, or, as he uses, Cloudflare Tunnels. He also recommends n8n's roughly two-hour YouTube course and leaning on Claude or Gemini to debug nodes.

2025-12-16 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2025-12-16 · open the original ↗

• AI fails society through bad models of people, not algorithms. Pentland says the Soviet Union ran its economy on early optimization AI and was undone by lied-about numbers, changing conditions and rare events, which he says is thought to have been a major contributor to its collapse. He sees the same failures in expert systems built for the average and in backward-looking machine learning, and answers "absolutely" when the host asks whether today's system could fall apart the same way.
• People are social foragers, and big voices drive polarization. Pentland cites Daniel Kahneman's estimate that about 95% of what we do comes from the culture around us. He says his group has shown that follower-based social media amplifies extreme voices and causes polarization, and that traders handling $10 million a day who watched other traders reliably beat those following equations alone.
• AI should direct attention and advise, never decide. His examples are on-premises "AI buddies" that read the manuals and track what colleagues do, which he says a good team can build in days, and the open-source deliberation.io, which visualizes everyone's views without big voices and uses an AI mediator. He reports people liked its AI summaries twice as much as human ones, and that Washington, DC residents wanted personal AI to "defend them against the city."
• Communities need data cooperatives to counter corporate data concentration. Pentland cites Xi Jinping calling data a means of production alongside capital and labor, and points to agricultural banking cooperatives that broke the Eastern banks' monopoly, built much of the US grid and became credit unions. Personal agents that share questions and answers rather than personal data, as in the loyal agents project with Consumer Reports and Stanford, could let members compete with firms like Walmart.
• Govern AI through open audit trails and standards, not pre-approval. Pentland argues that AI queries cross borders, so rules must be globally consistent, and proposes logging what AI does, to whom and with what data, as is done with money, with encryption or knowledge proofs against faked logs. He offers the UL testing cooperative as the model, says European regulators at the Club de Madrid called this "too Anglo-Saxon", and cites a California law requiring big labs to publish their safety practices.

2025-12-16 · open the original ↗

• Ives estimates Gemini's success adds about $56 a share to Google's stock. He says this validates his earlier bullish call on Google made when "London cab drivers" were still bearish.
• He still believes OpenAI would command a $1 trillion market cap if it went public, because it's building a "full stack" AI business beyond ChatGPT. He frames Gemini's rise and OpenAI's value as complementary rather than one coming at the other's expense.
• He argues the AI revolution needs many competing models, including ones from China, which he expects to get cheaper over time. This reflects his own view rather than a stated industry consensus.

2025-12-16 · open the original ↗

• "Arab slave trade" should be called the Muslim slave trade. He argues it included Indian Muslims, Berbers, Ottoman Turks, Persians and Central Asians, and that labelling it Arab is "kind of racist". He takes up the topic in defence of another creator being attacked for raising it.
• It lasted far longer than the transatlantic trade. He dates the transatlantic trade from the late 1400s to the late 1800s, and the Muslim slave trade from about the 600s to 1962, when he says Saudi Arabia finally outlawed slavery. He dismisses the view that the transatlantic trade was worse.
• He frames slaving as universal, not a sign of backwardness. He calls the medieval Islamic caliphates the most advanced society on earth, drawing scholars to Baghdad from Europe, India and China, and says slave trading was simply the easiest way to get rich, as it was for the Venetians and Vikings.

2025-12-16 · open the original ↗

• Intercal, the first esolang, rejects programs for bad manners. The clip, word for word from Hanselman's interview with Daniel Temkin, dates Intercal to 1972 as a parody that still comments on coding: too few PLEASE statements and it finds you rude, too many and it finds you simpering, and either way it ignores the whole program, with the threshold originally undocumented. Temkin says Don Woods recently found the long-lost original compiler code, put online earlier this year.

2025-12-15 · open the original ↗

• Making an agent deterministic often turns it into a Python script. Li, walking back his tweet that agents are "just Python scripts" as too scathing, says he is very pro-agent. When he wants something done exactly his way, though, an LLM's answer space is too wide, and at large scale even a 5% chance of a weird answer can be costly.
• Let the LLM decide whether to trigger; keep execution deterministic. Li's recent lesson is that the LLM should make the decision to run a workflow and the workflow itself should run deterministically, which he finds a good balance. The host adds that lowering temperature and frameworks for testing many responses are other ways to rein in variability.

2025-12-15 · open the original ↗

• Ives says his Palantir call, made around $10 a share, was built on knowing CEO Alex Karp and betting on leadership rather than on metrics alone. He says the call drew hate mail from institutional investors who called it "one of the worst calls" they'd seen.
• He credits Satya Nadella with transforming Microsoft from a skepticism-laden internal pick in 2014 into a cloud and AI leader. He calls Nadella underpriced and "perhaps the best CEO there is," crediting him with recognizing OpenAI's potential as early as 2012–2013.

2025-12-14 · open the original ↗

• Automating a workflow fails if nobody involved has mastered it. Li describes a side project, launched around the GPT image release, to auto-generate ads for founders to run on Meta's ad manager. It produced ads, but business owners told him they were bad, and he had no eye for good creative despite knowing the OpenAI APIs and Python.
• Do the work manually first, or partner with someone who can. Li concludes that the right setup is a person with intuition for what is good in the loop, either someone with both domain and tech skills or a partnership, and that you should prove you can get a good result by hand before automating. His ad-industry background in measurement, he notes, did not transfer to producing creative, unlike forecasting and time-series work he is confident automating.

2025-12-13 · open the original ↗

• Bannon went quiet on the Epstein files as release neared. He notes Steve Bannon had called for a special prosecutor, then fell silent around August or September. Photos show the two together, and he says Bannon filmed Epstein for 15 hours for a documentary that never surfaced, a working relationship dating to 2019 or earlier. He stresses he is accusing Bannon of nothing.
• The released photos show guilt by association, not wrongdoing. He walks through images of Woody Allen, Larry Summers, Richard Branson, Prince Andrew with Bill Gates, Bill Clinton beside Ghislaine Maxwell, and the current president at a party. He says nothing incriminating or lewd appears in any of them so far.