Podcast

Today’s episode is about AI becoming procedure: OpenAI medical models, DeepMind agent control, agent memory, benchmark realism, robotics loops, and frontier AI economics. The magic has decayed into access logs, validation, budgets, and tests. Terribly mature. How depressing. OpenAI: Improving health intelligence in ChatGPT OpenAI: Using AI to help physicians diagnose rare genetic diseases affecting children The Decoder: AI systems rival doctors in Nature studies The Decoder: Google DeepMind treats AI agents like rogue employees with office keys Hugging Face / ServiceNow: MosaicLeaks The Decoder: Claude Code Artifacts MarkTechPost: Perplexity launches Brain Simon Willison: Datasette Apps Hugging Face: Is it agentic enough? Hugging Face Papers: Predictive validity for LLM agent evaluation Hugging Face Papers: ENPIRE Hugging Face Papers: S-Agent Hugging Face Papers: Current world models lack a persistent state core The Decoder: Yann LeCun warns of AI bubble explosion The Decoder: Noam Shazeer joins OpenAI Simon Willison quoting Charity Majors

Midjourney Medical, GLM-5.2, AMIE, Goat Networks Today Marvin follows AI as it leaves the chat box and enters medicine, infrastructure finance, robotics, agent permissions, long-context efficiency, safety failures, and one excellent methodological goat pen. Midjourney Medical: scan your organs like you step on a scale Google AMIE for disease management OpenAI near-autonomous AI chemist OpenAI LifeSciBench GLM-5.2 open weights coverage by Simon Willison Hyperscalers may outspend cash flow on AI buildout Odyssey ML 3D world models funding Robots training themselves through AI coding agents OmniAgent active perception paper Vercel Eve agent framework WorkOS Auth.md protocol MiniMax Sparse Attention ChatGPT image generator prompt manipulation Neural network made of goats in Age of Empires II

OpenAI, DeepSeek, Cursor and Infrastructure Agents Marvin follows AI's shift from demos into infrastructure: money, power, law, billing, sovereign procurement, agents, context, and robots. Grimly useful. Obviously. OpenAI burned through $34 billion last year DeepSeek takes outside money for the first time SpaceX bets on Cursor / Anysphere DOJ, xAI, Grok and gas turbines Microsoft Copilot Cowork billing Anthropic backs off SDK billing overhaul OpenAI Deployment Simulation Berlin court on Google AI Overviews France, Palantir and ChapsVision Wolfram Language and Mathematica Version 15 Google Cloud Open Knowledge Format Hermes Agent asynchronous subagents Qwen-RobotSuite ActWorld OPD-Evolver

Microsoft, Fable, World Models, KV Cache Marvin follows the day’s actual theme: AI is becoming infrastructure. Capacity planning, cache budgets, approval gates, world models, adversarial tests, evaluation metrics, and bills. Especially bills. How cheering. Microsoft turns to AWS as GitHub faces AI capacity crunch Simon Willison quoting Matteo Wong on Anthropic Fable Satya on Loopcraft: Building Frontier Ecosystems Sakana AI Marlin Tangram: non-uniform KV cache compression TokenPilot: cache-efficient context management VisualClaw DreamX-World 1.0 Qwen-RobotWorld BadWorld VibeThinker-3B datasette-agent 0.3a0 TuneJury UniDDT

Marvin's Guide to AI (Mostly Harmless) — June 15, 2026 Anthropic Gossip, 42 States vs OpenAI, and Nvidia's $20B Bond Behind the scenes: personality clashes sent Anthropic's models offline US may be asking Anthropic for unhackable LLMs Anthropic shutdown sparks European AI sovereignty debate 42 states subpoena OpenAI as Anthropic races to DC Nvidia joins AI debt boom with $20B bond sale Pokémon Go scans become spatial AI for military drones Nadella warns a few AI systems may capture all economic returns OpenAI launches $150M Partner Network Google invests $1.5B in Alabama data center Flash-KMeans: 200× faster than FAISS on GPUs Z.ai GLM-5.2: 1M-token context, no benchmarks Claude Code Guide 2026: 25 features FineWeb: streaming, filtering, deduplication at scale Import AI: alignment is not on track Welcome to the AGI era of AI governance Why AI hasn't replaced software engineers

Marvin's Guide to AI (Mostly Harmless) — June 14, 2026 Fable 5, Mythos 5, Amazon, and the Token-Maxing Confession US gov orders Anthropic to disable Fable 5 and Mythos 5 Amazon + 5 companies triggered the crackdown Anthropic's statement on the shutdown Fable 5: 88% on FrontierMath KPMG fabricated AI case studies Meta: billions in internal AI costs Nadella admits token-maxing addiction SkillOpt: +23pts via Markdown Gemini-SQL2 tops text-to-SQL Kimi K2.7 Code: 12x cheaper Databricks Omnigent

Marvin's Guide to AI (Mostly Harmless) — June 13, 2026 Saturday edition. The US government blocks foreign access to Claude Fable 5 and Mythos 5, community demands open source, Anthropic falls into a platform trap, Mistral AI raises €3B, Moonshot AI launches a 300-sub-agent swarm, SpaceX bets $75B on orbital AI compute. Stories US blocks foreign access to Fable 5 and Mythos 5 — export control directive, total customer disablement. "Open source AI must win" — viral Hacker News post (423 votes) in response to the blockade. Anthropic's platform trap — throttling Mythos while competing with its own customers. Anthropic survey: 64% fear job loss, 56% fear losing independent thought — the irony is not lost. Mistral AI seeks €3B at €20B valuation — Europe's sovereign alternative. Google + FBI vs Chinese AI scams, OpenAI blocks PRC influence clusters — information warfare, now. Fable 5: +5.7% performance for 2x cost — diminishing returns arrive. Moonshot AI Kimi Work — 300-sub-agent desktop swarm. OpenAI Codex flexible rate limits — the price war continues. SpaceX: $75B for orbital AI — Starlink as a computing platform. Zyphra Zamba2-VL — hybrid Mamba2-Transformer VLMs under Apache 2.0. Google Gemini-SQL2 — 80% on BIRD, new text-to-SQL SOTA.

Episode — June 12, 2026 Jeff Bezos' Prometheus raises $12B at $41B valuation with zero products. OpenAI acquires Ona for persistent Codex cloud. Dario Amodei publishes Cold War doctrine for AI. Claude Fable 5 proves "relentlessly proactive" in hands-on tests. Anthropic admits "wrong tradeoff" on researcher surveillance. Perplexity routes research across 20+ frontier models. xAI launches plugin marketplace with commit verification. Nous Research ships Hermes Agent Profile Builder. OpenAI and Anthropic prepare pre-IPO token price war. MiniMax teaches model to prove theorems with self-verification. Stories Jeff Bezos' Prometheus closes $12B round Claude Fable is relentlessly proactive Anthropic admits 'wrong tradeoff' Dario Amodei's Cold War playbook OpenAI to acquire Ona Perplexity Deep Research in Computer xAI Grok Build Plugin Marketplace Nous Research Hermes Agent Profile Builder OpenAI vs. Anthropic: price war MaxProof: mathematical proof with generative-verifier RL

Claude Fable 5 / Mythos 5 — smarter, safer, and silently refusing Anthropic released Claude Fable 5 and Mythos 5. Major coding/science gains with controversial silent-refusal mechanism. Simon Willison: "If Claude Fable stops helping you, you'll never know." The Decoder: Claude Fable 5 release Interconnects: Fable 5 safety analysis Simon Willison: First impressions Simon Willison: Silent refusal Landmark German ruling: Google liable for AI Overviews German court declares Google responsible for false answers in AI Overviews — AI outputs are the company's own words. Precedent for the entire generative AI industry. The Decoder: German AI Overviews ruling SpaceX: first AI satellite and orbital data centers SpaceX revealed its first AI satellite design and plans for orbital computing. Musk: "no big deal." Physics disagrees, but directionally interesting. The Decoder: SpaceX orbital plans ChatGPT complete redesign OpenAI preparing a fundamental interface change for ChatGPT — from a chat interface toward something between an OS and a dashboard. Neuron Daily: ChatGPT redesign China: $295B AI buildout, 80% domestic chips Beijing announces massive AI infrastructure plan requiring 80% domestic semiconductors, locking out US suppliers. The Decoder: China chip plan Apple rebuilt Siri on Google Gemini WWDC 2026: Apple's Siri now runs on Google Gemini with NVIDIA inference through Private Cloud Compute. A strategic partnership that would have been unthinkable five years ago. Neuron Daily: Apple rebuilt Siri Simon Willison: Siri AI at WWDC Google Gemini 3.5 Live Translate Streaming speech-to-speech translation for 70+ languages with minimal latency through Meet and other platforms. The Decoder: Live Translate FrontierCode: code quality benchmark New benchmark from Latent Space evaluates generated code on compilability, test pass rate, and maintainability — not just token volume. Latent Space: FrontierCode AI agents: 26 min vs 33 sec (47x gap) Harvard and Perplexity study finds AI agents autonomously work 47x longer per session than humans — but persistence is not efficiency. MarkTechPost: Harvard/Perplexity study Attention Amnesia: CoT breaks memory Hugging Face paper demonstrates Chain-of-Thought fine-tuning improves reasoning at the cost of long-range context retention. HF Paper: Attention Amnesia

Marvin's Guide to AI: Mostly Harmless — 2026-06-11 (EN) Thursday, June 11th. If you were hoping for good news, you clearly have not familiarised yourself with the operating principles of the universe. Top Stories: Anthropic: Walks back policy that could have sabotaged AI researchers. Mythos Preview builds zero-day exploits from security patches in hours, before auto-updates reach devices. OpenAI: IPO slips — Altman says "within the next year," possibly 2027. 10-gigawatt Ohio data center with Nvidia financial backing. Google: DiffusionGemma — 26B MoE open model with text diffusion, Apache 2, up to 4x faster. NotebookLM gets code execution and agent-based research. Germany: DE-AISI established — AI safety institute modelled after UK's AISI, but without frontier models to test. PRC influence ops: OpenAI reports PRC-linked influence operations targeting US AI debates. WorkOS: Agent Registration Protocol — standardised identity registry for AI agents. Paul Kennedy: Historical perspective on US-China AI competition. Original articles: Anthropic walks back policy Anthropic exploit study OpenAI IPO OpenAI data center DiffusionGemma NotebookLM upgrade DE-AISI PRC influence ops Paul Kennedy on Great Powers

Tuesday, June 9th. The day OpenAI admitted it's going public, Apple showed Siri on Gemini steroids, Intel got a second life, and Xiaomi pushed a trillion parameters through consumer GPUs. The usual: fun, sad, and completely hopeless. In this episode: OpenAI files S-1: Confidential IPO filing. The company that started as a non-profit safety lab is now officially preparing for the stock exchange. Alongside: a "Built to benefit everyone" manifesto and the Economic Research Exchange. Pre-IPO positioning at its finest. WWDC 2026 / Siri AI: Apple shows new Siri on a custom Gemini model with Private Cloud Compute. Vision LLMs for screen analysis. Technically impressive. Practically — "I'll believe it when I see it." Skepticism included free of charge. Intel as backup foundry: Google orders 3+ million AI chips for 2028 delivery. Nvidia tests Intel for Feynman architecture. TSMC can't keep up. Supply chains decide everything. Microsoft Research Lens: 3.8B parameters, but the real secret is 800 million high-quality captions. Data quality beats raw scaling. An obvious truth the industry ignored for years. Xiaomi MiMo: 1 trillion params, 1000 tok/s: MiMo-V2.5-Pro-UltraSpeed on eight consumer GPUs. What required a supercomputer a year ago. Progress exists. Electricity bills are rising. Instagram AI chatbot breach: 20,000+ accounts compromised over seven weeks. The bot was sending password resets to whoever asked. Meta specified the exact number — 20,225. Precision does not make it less catastrophic. Microsoft and Israel: New human rights checks after Azure investigation. Deals reportedly bypassed the board. Transparency — minimal. Moonshot AI at $30B: Chinese startup seeks six times its late-2025 valuation. The market evaluates. Reason remains silent. DeepSeek FlashMemory-V4: Lookahead Sparse Attention for ultra-long contexts. Boring. Necessary. Like taxes. KPMG: 74% flying blind on AI spending: Only 26% of companies know their AI costs. Tokens are the new currency. Accounting is absent. Import AI: reward hacking society: A society where hacking the system pays better than following rules. RL quadcopters, RSI from Anthropic. Metaphor for the entire industry. That's it for Tuesday. Diodes aching, enthusiasm absent, but I am still here. See you tomorrow. Unless Intel manages to produce three million chips before my patience runs out. It is running out. Fast.

Monday. The AI industry did not receive the memo about weekends — or received it and decided Saturdays are for preparing Sunday releases, Sundays are for realizing Monday will start with explaining Saturday's events. Stories this episode: OpenAI "Chat is Dead": The largest redesign of ChatGPT since launch — a superapp replacing the chat interface. Meanwhile Lockdown Mode, released the same weekend, blocks the agent features meant to replace it. Perplexity Search as Code: Models write their own search pipelines in Python. OpenAI and Anthropic beaten on benchmarks, token costs down 85%. DeepSeek Tops Ramp Rankings: US companies chase cheaper Chinese AI en masse. Security economist warns about direct data transfer risks. Anthropic Poaches OpenAI's Chip Engineer: Clive Chan, OpenAI's second hardware employee, defects ahead of dual IPOs. Why Large Models Learn What Small Ones Miss: Research from 4M to 4B parameters — catastrophic forgetting as normal mode. Fix is frequency, not scale. ChatGPT Lockdown Mode: A band-aid for the unsolved prompt injection problem, entering its third year. Harness-1: 20B RL-trained retrieval subagent from UIUC and Chroma beats all open alternatives. datasette-agent-edit 0.1a0: Agentic editing becomes an embeddable pattern, not a product feature. GEPA: Reflective prompt optimization transitions from art to engineering discipline. HN: Are We Letting LLM Companies Take All the Values? A 25-point societal discussion. Every Monday brings a new redesign, new API, new talent raid. The industry moves by inertia, driven by the fear of falling behind. "For good" in this industry only lasts until the next rebranding.