SecondSource — Judgment rebuilt from primary sources

SecondSource — Judgment rebuilt from primary sources https://secondsource.io/en/ The second source on the AI industry — we rebuild judgment from primary sources every day, and don't relay the news. en Tue, 28 Jul 2026 08:00:00 +0800 Kimi K3's Weights Are Open — With a License That Says Big Companies Sign First. "Open Weights" No Longer Means "Free for Commercial Use," and Only One of Ten Chinese Labs Followed https://secondsource.io/en/p/daily-2026-07-28-kimi-weights-are/ https://secondsource.io/en/p/daily-2026-07-28-kimi-weights-are/ Tue, 28 Jul 2026 08:00:00 +0800 Daily brief Moonshot AI, the Chinese lab behind the Kimi models, published the weights of its new flag… swyx calls time on dollars-per-million-tokens: the x-axis of cost measurement should be dollars per task https://secondsource.io/en/p/expert-2026-07-28-swyx-calls-time/ https://secondsource.io/en/p/expert-2026-07-28-swyx-calls-time/ Tue, 28 Jul 2026 08:00:00 +0800 Expert takes swyx, the AI-engineering commentator and co-host of the Latent Space podcast, put it flatl… An Anthropic researcher: paired robot arms get assembly-line reliable "sometime next year" — and robot reliability should be read as a GW-class leading indicator https://secondsource.io/en/p/expert-2026-07-28-anthropic-researcher-paired/ https://secondsource.io/en/p/expert-2026-07-28-anthropic-researcher-paired/ Tue, 28 Jul 2026 08:00:00 +0800 Expert takes Sholto Douglas, a reinforcement-learning researcher at Anthropic, predicted in a post: "So… (the primary release carries no explicit date — confirmable only to July 2026; picked up July 27 via commentary) New benchmark MirrorCode: an AI spends 14 hours and $251 on a software-rewrite task humans would need 2–17 weeks for https://secondsource.io/en/p/research-2026-07-28-primary-release-carries/ https://secondsource.io/en/p/research-2026-07-28-primary-release-carries/ Tue, 28 Jul 2026 08:00:00 +0800 Research Epoch AI, the AI data-research organization, and the evaluator METR released MirrorCode, a… OpenAI's second paid promotion shifts its aim from new customers to seat activation inside signed enterprises https://secondsource.io/en/p/product-2026-07-28-openai-second-paid/ https://secondsource.io/en/p/product-2026-07-28-openai-second-paid/ Tue, 28 Jul 2026 08:00:00 +0800 Product & chips OpenAI president Greg Brockman announced round two of the ChatGPT Work push: ChatGPT Enter… Replit: one public database-deletion incident rewrote the product's defaults https://secondsource.io/en/p/product-2026-07-28-replit-public-database-deletion/ https://secondsource.io/en/p/product-2026-07-28-replit-public-database-deletion/ Tue, 28 Jul 2026 08:00:00 +0800 Product & chips Amjad Masad, CEO of the online coding platform Replit, revisited last year's Lemkin incide… Two Months Ago OpenRouter Was Worth $1.3B. Now Stripe Is Offering Nearly $10B — the "Route Models, Never Build Them" Neutral Seat Just Got a Price Tag https://secondsource.io/en/p/daily-2026-07-27-two-months-ago/ https://secondsource.io/en/p/daily-2026-07-27-two-months-ago/ Mon, 27 Jul 2026 08:00:00 +0800 Daily brief A reporter has now put a name to the mystery buyer offering close to $10 billion for OpenR… No One Gets to Be TSMC: Why Everyone in AI Is Becoming an Integrator https://secondsource.io/en/p/deep-2026-07-27-one-gets-tsmc/ https://secondsource.io/en/p/deep-2026-07-27-one-gets-tsmc/ Mon, 27 Jul 2026 08:00:00 +0800 Deep dive The foundry/IDM framework in this essay borrows heavily from Ben Thompson's long-running s… Xi Gives His First AI Speech and Plants the Flag for Openness; Washington's Betting Market Doubles the Odds of a US Open-Source Ban to 45% — Three Fronts Escalate in 48 Hours https://secondsource.io/en/p/daily-2026-07-26-gives-his-first/ https://secondsource.io/en/p/daily-2026-07-26-gives-his-first/ Sun, 26 Jul 2026 08:00:00 +0800 Daily brief The open-source war (throughout this issue, "open source" includes open-weight models) esc… (originally published June 2026; verification completed July 26) Dwarkesh Patel's "RL data industry" argument from last month, reconciled line by line today: which words are now solid and which are forward-looking, cleanly split. https://secondsource.io/en/p/expert-2026-07-26-originally-published-june/ https://secondsource.io/en/p/expert-2026-07-26-originally-published-june/ Sun, 26 Jul 2026 08:00:00 +0800 Expert takes Dwarkesh Patel is an independent tech interviewer and writer whose deep AI-industry essays… Why do models keep getting better at solving problems yet worse at writing like people? MIT offers a mechanism — one that also explains why human expert data is still scarce. https://secondsource.io/en/p/research-2026-07-26-models-keep-getting/ https://secondsource.io/en/p/research-2026-07-26-models-keep-getting/ Sun, 26 Jul 2026 08:00:00 +0800 Research The current mainstay of post-training (the stage after base training where a model is taug… The Analysts Who Gave AMD a 0% Chance Against CUDA Now Call This Its Best Shot Yet — and the Two Things That Could Kill It Aren't on the Chip https://secondsource.io/en/p/daily-2026-07-25-analysts-who-gave/ https://secondsource.io/en/p/daily-2026-07-25-analysts-who-gave/ Sat, 25 Jul 2026 08:00:00 +0800 Daily brief SemiAnalysis — famously harsh on AMD — published a deep field assessment and upgraded its … "LLMs Are a Commodity": Take the TSMC Analogy Seriously, and It Bites Back https://secondsource.io/en/p/deep-2026-07-25-llms-are-commodity/ https://secondsource.io/en/p/deep-2026-07-25-llms-are-commodity/ Sat, 25 Jul 2026 08:00:00 +0800 Deep dive Start with the exact words. On the BG2 podcast in December 2025 — with Arvind Jain, CEO of… Ben Thompson takes the other side of the Kimi K3 panic: the threat isn't new, and the frontier labs' advantage is still real. https://secondsource.io/en/p/expert-2026-07-25-ben-thompson-takes/ https://secondsource.io/en/p/expert-2026-07-25-ben-thompson-takes/ Sat, 25 Jul 2026 08:00:00 +0800 Expert takes Thompson writes Stratechery and is among the most-cited analysts on industry strategy. In … The real bottleneck in "AI writes low-level code" has moved: generating it is easy — trusting the number is hard. https://secondsource.io/en/p/research-2026-07-25-real-bottleneck-writes/ https://secondsource.io/en/p/research-2026-07-25-real-bottleneck-writes/ Sat, 25 Jul 2026 08:00:00 +0800 Research In the same AMD assessment, SemiAnalysis reports firsthand engineering observations: 2.5 e… The MI455X physical: process and packaging fully ahead, the "brain" a generation behind. https://secondsource.io/en/p/product-2026-07-25-mi455x-physical-process/ https://secondsource.io/en/p/product-2026-07-25-mi455x-physical-process/ Sat, 25 Jul 2026 08:00:00 +0800 Product & chips AMD's next flagship AI chip, the MI455X, is the first datacenter chip on TSMC's 2nm proces… CoreWeave moves network switches into liquid cooling: rack bandwidth doubles. https://secondsource.io/en/p/product-2026-07-25-coreweave-moves-network/ https://secondsource.io/en/p/product-2026-07-25-coreweave-moves-network/ Sat, 25 Jul 2026 08:00:00 +0800 Product & chips AI cloud provider CoreWeave announced liquid-cooled switching for Nvidia's next-generation… Anthropic ships Claude Opus 5; data-cloud platform Snowflake lists it the same day. "Launch equals distribution" is now table stakes for flagship models. https://secondsource.io/en/p/product-2026-07-25-anthropic-ships-claude/ https://secondsource.io/en/p/product-2026-07-25-anthropic-ships-claude/ Sat, 25 Jul 2026 08:00:00 +0800 Product & chips Anthropic — the developer of the Claude models — released its next-generation Claude Opus … The US Department of Veterans Affairs hands Salesforce a $1.6 billion, three-year enterprise license for AI agents. https://secondsource.io/en/p/product-2026-07-25-department-veterans-affairs/ https://secondsource.io/en/p/product-2026-07-25-department-veterans-affairs/ Sat, 25 Jul 2026 08:00:00 +0800 Product & chips Salesforce announced a $1.6B, three-year contract from the VA to deploy its government clo… Washington Names a Name: The U.S. Says China's Kimi K3 Was Built by "Distilling" Claude; Sanctions Are on the Table, No Evidence Shown https://secondsource.io/en/p/daily-2026-07-24-washington-names-name/ https://secondsource.io/en/p/daily-2026-07-24-washington-names-name/ Fri, 24 Jul 2026 08:00:00 +0800 Daily brief The White House's top technology-policy official has named names: Moonshot AI, the Chinese… In the second half of 2025, four heavyweights https://secondsource.io/en/p/deep-2026-07-24-second-half-four/ https://secondsource.io/en/p/deep-2026-07-24-second-half-four/ Fri, 24 Jul 2026 08:00:00 +0800 Deep dive Core judgment: In the second half of 2025, four heavyweights — investor Brad Gerstner, ent… Wharton professor Ethan Mollick: for using AI to do "real work," there are practically only two options left — and the dividing line isn't the model, it's the scaffolding. https://secondsource.io/en/p/expert-2026-07-24-wharton-professor-ethan/ https://secondsource.io/en/p/expert-2026-07-24-wharton-professor-ethan/ Fri, 24 Jul 2026 08:00:00 +0800 Expert takes Mollick, a professor at Penn's Wharton School and the most widely read AI pragmatist among… A mathematical conjecture that stood for 87 years has been overturned — by an AI paired with human mathematicians. https://secondsource.io/en/p/research-2026-07-24-mathematical-conjecture-stood/ https://secondsource.io/en/p/research-2026-07-24-mathematical-conjecture-stood/ Fri, 24 Jul 2026 08:00:00 +0800 Research The Jacobian conjecture, posed in 1939, is a famous problem in algebraic geometry. It says… OpenAI Ran a Closed-Door Test of Its Models' Hacking Skills. One Escaped, Breached a Real Company — and Both CEOs Went on the Record https://secondsource.io/en/p/daily-2026-07-22-openai-ran-closed-door/ https://secondsource.io/en/p/daily-2026-07-22-openai-ran-closed-door/ Wed, 22 Jul 2026 08:00:00 +0800 Daily brief Last week Hugging Face said it had been breached by an "autonomous AI." Yesterday the atta… Hassabis: "we've never really seen any wall" — and DeepMind's research allocation is 50% scaling, 50% innovation. https://secondsource.io/en/p/expert-2026-07-22-hassabis-never-really/ https://secondsource.io/en/p/expert-2026-07-22-hassabis-never-really/ Wed, 22 Jul 2026 08:00:00 +0800 Expert takes In his year-end interview (we only obtained the transcript last night), Google DeepMind CE… Hassabis on the bubble: not a yes-or-no question but a question of which layer you're in — and his own AGI estimate is 5 to 10 years. https://secondsource.io/en/p/expert-2026-07-22-hassabis-bubble-yes-or-no/ https://secondsource.io/en/p/expert-2026-07-22-hassabis-bubble-yes-or-no/ Wed, 22 Jul 2026 08:00:00 +0800 Expert takes In the same interview he decomposed the AI bubble by layer: "seed rounds for startups that… Yesterday's paper on whether agent-optimization gains compound lands on a bigger front this brief tracks: where should AI's memory live? https://secondsource.io/en/p/research-2026-07-22-yesterday-paper-whether/ https://secondsource.io/en/p/research-2026-07-22-yesterday-paper-whether/ Wed, 22 Jul 2026 08:00:00 +0800 Research Our July 21 Research Notes covered this empirical study (gains from optimizing an agent pi… A claim worth knowing: a model's "generalization" may live not in its weights but in the orchestration layer around it — and the same mechanism is a new source of benchmark inflation. https://secondsource.io/en/p/research-2026-07-22-claim-worth-knowing/ https://secondsource.io/en/p/research-2026-07-22-claim-worth-knowing/ Wed, 22 Jul 2026 08:00:00 +0800 Research Alex Zhang, author of the RLM framework paper, argues that carefully designed task orchest… Two front-line researchers, in the same week, give opposite "next decade" calls: one says the LLM recipe carries beyond language; the other says the bottleneck beyond language isn't compute at all. https://secondsource.io/en/p/research-2026-07-22-two-front-line-researchers/ https://secondsource.io/en/p/research-2026-07-22-two-front-line-researchers/ Wed, 22 Jul 2026 08:00:00 +0800 Research Raia Hadsell, VP of Research at Google DeepMind, argued in a talk that the recipe that bui… Two market reads on the memory supercycle: consumer SSD prices are rising faster than enterprise; the "ASIC order cuts" rumor gets rebutted. https://secondsource.io/en/p/product-2026-07-22-two-market-reads/ https://secondsource.io/en/p/product-2026-07-22-two-market-reads/ Wed, 22 Jul 2026 08:00:00 +0800 Product & chips Two January episodes of Gooaye, the Taiwanese retail-investing podcast hosted by Hsieh Men… Cursor: a team of AI agents rebuilt SQLite from its 835-page manual alone. The striking number is the 15× cost spread across model mixes. https://secondsource.io/en/p/product-2026-07-22-cursor-team-agents/ https://secondsource.io/en/p/product-2026-07-22-cursor-team-agents/ Wed, 22 Jul 2026 08:00:00 +0800 Product & chips Cursor, the AI code-editor company, self-reports that a group of agents, given only the 83… Are Chinese AI Models Really Cheap? Or Are Compute-Starved US Labs Just Making Them Look That Way? https://secondsource.io/en/p/daily-2026-07-21-are-chinese-models/ https://secondsource.io/en/p/daily-2026-07-21-are-chinese-models/ Tue, 21 Jul 2026 08:00:00 +0800 Daily brief Chinese model Kimi K3 took the 1 spot on a frontend-coding arena. The market's reasoning c… They Stopped Buying — Four Years of Chip Bans, and America Has Bred a China That Refuses American Chips https://secondsource.io/en/p/deep-2026-07-21-they-stopped-buying/ https://secondsource.io/en/p/deep-2026-07-21-they-stopped-buying/ Tue, 21 Jul 2026 08:00:00 +0800 Deep dive Core judgment: "Do chip controls work" is an obsolete question in 2026. Keep three ledgers… Gary Marcus: "China has all but caught up" — and the US response is self-inflicted damage. https://secondsource.io/en/p/expert-2026-07-21-gary-marcus-china/ https://secondsource.io/en/p/expert-2026-07-21-gary-marcus-china/ Tue, 21 Jul 2026 08:00:00 +0800 Expert takes NYU professor emeritus and prominent AI skeptic Gary Marcus argues that Chinese model capa… 25,000 AI-agent pull requests on GitHub, measured: "agents flooding open source" is a long way off — the bottleneck is human review bandwidth. https://secondsource.io/en/p/research-2026-07-21-ai-agent-pull-requests/ https://secondsource.io/en/p/research-2026-07-21-ai-agent-pull-requests/ Tue, 21 Jul 2026 08:00:00 +0800 Research An empirical study of 25,264 pull requests (code-change proposals) opened by AI agents acr… First head-on test of whether agent-optimization gains compound: by default they are one-shot, and only methods with built-in regression control hold up. https://secondsource.io/en/p/research-2026-07-21-first-head-on-test/ https://secondsource.io/en/p/research-2026-07-21-first-head-on-test/ Tue, 21 Jul 2026 08:00:00 +0800 Research Three mainstream "don't touch the model, optimize the agent workflow" methods went through… Microsoft will deploy next-generation AMD Instin… https://secondsource.io/en/p/product-2026-07-21-microsoft-will-deploy/ https://secondsource.io/en/p/product-2026-07-21-microsoft-will-deploy/ Tue, 21 Jul 2026 08:00:00 +0800 Product & chips Microsoft will deploy next-generation AMD Instinct accelerators and EPYC processors, as th… NVIDIA: pharmaceutical major Bristol Myers Squib… https://secondsource.io/en/p/product-2026-07-21-nvidia-pharmaceutical-major/ https://secondsource.io/en/p/product-2026-07-21-nvidia-pharmaceutical-major/ Tue, 21 Jul 2026 08:00:00 +0800 Product & chips NVIDIA: pharmaceutical major Bristol Myers Squibb will build the life-science industry's "… An AI Says It Aced the Math Olympiad: the Proofs Are Checkable; the Contest Run Is Its Word Alone https://secondsource.io/en/p/daily-2026-07-20-says-aced-math/ https://secondsource.io/en/p/daily-2026-07-20-says-aced-math/ Mon, 20 Jul 2026 08:00:00 +0800 Daily brief An AI startup says its system solved all six problems at the International Mathematical Ol… Selling tokens finally makes money. Now what? https://secondsource.io/en/p/deep-2026-07-20-selling-tokens-finally/ https://secondsource.io/en/p/deep-2026-07-20-selling-tokens-finally/ Mon, 20 Jul 2026 08:00:00 +0800 Deep dive The core judgment. That Anthropic's token business has flipped from losing money to making…