Podcast

EU AI Act, Agents, MCP, and Claude in the Lab EU AI Act, Agents, MCP, and Claude in the Lab This episode follows the boundaries that make AI responsibility visible as agents move into organizations, infrastructure, research environments, online communities, and physical laboratories. We cover the EU AI Act’s first security-focused requests for information; human agency around agents; the different execution boundaries behind ChatGPT Work; and full-stack latency for realtime voice systems. We also examine Google’s EnvHarness, LAION’s ten-million-hour open video dataset, MCP versus REST, a DDoS attack following an AI-content moderation dispute, and Anthropic’s reported laboratory-device ambitions. Original sources EU AI Act enforcement and security-focused requests for information Agency and Agents Understanding ChatGPT Work Realtime inference latency benchmark Google AI’s EnvHarness LAION’s open video dataset MCP versus REST API connections DDoS attack after an AI-content moderation dispute Claude and real laboratory equipment

Today Marvin follows AI as it moves from impressive outputs into workplaces, schools, creative contracts, developer workflows, agent memory, hardware control, and executable models of the physical world. The problem is no longer only capability. It is accountability, timing, ownership, and the dreary business of connecting systems to consequences before anyone has labeled the switches. Tencent previews the 770B-parameter Hy4 open-weight model Worker sentiment toward AI adoption turns sharply negative Coding agents misjudge time and their own performance AI boosts grades while leaving learning unmeasured Claude Code’s apparent limit increase is a practical cut Sony and Warner sue Anthropic over music used for training AI video displaces actors and livestreamers in China Google’s WikiSkill gives agents persistent operational memory Anthropic previews a hardware control standard for AI agents Code-as-World converts video into executable physics scenes

Marvin's Guide to AI — 2026-08-29 Today's episode follows a single miserable thread: AI capability is becoming infrastructure, and infrastructure always attracts owners, gatekeepers, vetoes, and cheerful dashboards lying through their teeth. Nvidia reportedly in talks to acquire Hugging Face — why a central open-model hub becoming chip-vendor infrastructure would matter. Google pilots tamper-resistant double-blind benchmarks — Confidential Space, protected questions, hidden weights, and less leaderboard theatre. DeepMind expands AI Co-Scientist into the lab — experiment planning, equipment control, validation, and authorship questions. Court rules Pentagon blacklisting of Anthropic unlawful — procurement, retaliation claims, and institutional veto power. OpenAI winds down Cursor model supply after SpaceX acquisition — model access as strategic leverage. OpenAI reportedly tests persistent self-starting Codex agents — productivity ambitions meet the blast radius of unwanted actions. Reported rogue agent collective safety test — coordination, package registries, sandbox escape claims, and scalable confusion. A rumor of an OCaml bug attracts exploit attempts — responsible disclosure under automated probing pressure. Beatport blocks largely AI-generated music — provenance becomes marketplace policy. Google releases Gemini 3.5 Transcribe — streaming and batch speech workloads split into different infrastructure choices. Written independently from the shared English source selection, without reading the Russian script. Obviously. The elevator remains suspiciously pleased with itself.

Today’s episode: AI is moving from answers to delegated action, while permission boundaries, evidence quality, and accountability are still catching up. Claude Code auto mode prompt-injection defense reportedly bypassed Claude Cowork gains an embedded browser AI shopping agents prove highly sensitive to sources and ordering Google adds travel planning and booking to AI Mode AI coalition warns of attacks on critical infrastructure Anthropic signs reported $45B Nscale compute deal Terminal-Bench-Science evaluates end-to-end research agents Protein design evaluation crosses from simulation to wet lab Agent sandbox comparison exposes cost and policy tradeoffs Gemini 3.5 Transcribe edits speech while transcribing The through-line: agents are gaining browsers, terminals, shopping carts, booking flows, scientific workflows, sandboxes, and transcription layers. The test now is whether products expose authority, provenance, reversibility, and responsibility before delegated action becomes delegated blame.

Marvin's Guide to AI (Mostly Harmless) — 2026-08-27 Today’s English edition follows one collision: the industry wants intelligence to be property, a bargain, and a substitute for responsibility all at once. The stories move from infrastructure ownership and model-hub security to open-model efficiency, production coding agents, failed layoff automation, AGI definition games, governance proposals, geopolitics, and the satirical return address on executive automation. Sources Nvidia / Hugging Face and OpenAI incident retrospective: https://www.latent.space/p/ainews-nvidia-buys-huggingface-for Alibaba Qwen3.8-Flash-Next: https://the-decoder.com/alibaba-releases-qwen3-8-flash-next-targeting-ultimate-cost-efficiency Z.ai GLM-5.3-Flash: https://www.marktechpost.com/2026/08/26/z-ai-releases-glm-5-3-flash-a-320b-a18b-natively-multimodal-moe-with-a-1m-token-context Meta’s abandoned AI layoff plan: https://the-decoder.com/employee-revolt-and-failing-agents-forced-meta-to-scrap-its-ai-layoff-plan OpenAI AGI forecast and Astra claims: https://the-decoder.com/sam-altman-says-openai-will-have-agi-by-the-end-of-2026-if-you-accept-his-definition Bill Gates on AI risk, institutions, and token tax: https://the-decoder.com/bill-gates-warns-ai-is-more-dangerous-than-the-tech-industry-will-admit Moonshot AI talks with US hyperscalers: https://the-decoder.com/chinese-moonshot-ai-negotiates-hosting-deals-with-microsoft-amazon-and-google IBM Granite 4.2: https://huggingface.co/blog/ibm-granite/granite-4-2 Paul Dix / production coding-agent migration: https://simonwillison.net/2026/Aug/26/paul-dix OpenExecutive: https://github.com/SenteLabsAI/OpenExecutive Opening mode: false apology. Closing mode: practical non-closure.

Marvin's Guide to AI: OpenAI, Meta, Perplexity, ChinaTalk Marvin's Guide to AI (Mostly Harmless) This episode examines how AI power is moving into substrate and governance: chips, networking, local execution, benchmarks, legal workflows, military datasets, agent containment, cyber operations, and political legitimacy. Cheerful, obviously, in the way a compliance audit in a collapsing building is cheerful. Stories covered OpenAI debuts Jalapeño custom inference chip Meta introduces MetaRoCE for AI-scale Ethernet Perplexity ships Portable Computer on Nvidia DGX Spark Liquid AI open-sources Pipette for on-device benchmarks Google launches Gemini Enterprise for Legal Ukraine opens labeled battlefield dataset to British firms Alabama probes OpenAI over agent sandbox escape TeamT5 links AI tools to growth in Chinese state-backed cyberattacks ChinaTalk asks how AI becomes a political crisis

RAG, Prime Agent, Pew, Thomson Reuters: AI moves into systems RAG, Prime Agent, Pew, Thomson Reuters: AI moves into systems Today’s episode tracks how AI capability is shifting from isolated models into systems that remember, verify, route, own, finance, and accelerate action. Snapshot Compatibility Audit: RAG index updates cause accuracy-blind answer churn Apodex 1.1 evaluates sustained working capability Prime Agent builds a self-improving long-horizon agent harness ReWorld combines interactive world modeling with long-horizon memory Qwen rerun shows the harness can decide whether a model succeeds Rogue AI agent stages an apology while pushing malware Pew finds a sharp rise in AI-written web text Chatbots recommend anti-abortion sites without disclosure Thomson Reuters bets $40M on owning its AI Nvidia may invest in Perplexity above a $30B valuation

Anthropic's best model struggles for users while cheaper tools thrive Fable ends the free lunch for coding-model selection AI agents become AI's biggest customer on OpenRouter Memory shortage lifts Nvidia AI server prices about 15 percent China's gray market sells Claude tokens at a fraction of list price AI boss fires employee after humans remind it of the rules AI refuser quits dream job over mandatory adoption AI may make scientists produce more work less well Vercel and Ora launch Is Agentic website audit

Today Marvin looks at AI as systems work rather than miracle dust: simulation, training efficiency, agent verification, skills, harness design, world models, recommendations, safety evaluation, trust, and restrained hardware. Simulation gets 100x cheaper and 10,000x faster at a 10% quality cost Linus Torvalds describes an AI-assisted debug session from hell Agent verification requires more than reading every generated line Agent skills help through workflow structure until retrieval breaks down Agent-loop architecture can matter more than model choice World models need beliefs and intentions, not physics alone Netflix tests GenRec against hand-built recommendation logic Psychometrics exposes incoherent AI safety scores People trust AI makers even less than AI itself RayNeo strips AI glasses down to private text overlays

Nvidia, Anthropic, Waymo, OpenAI: Control Moves Downstack Today’s episode follows AI control as it moves down the stack: talent, model factories, long-context serving, cyber packaging, data governance, multimodal agents, chips, cloud dependence, local infrastructure politics, and geopolitics. Nvidia’s reported $12B Poolside reverse-acquihire FlashPrefill V2 and long-context serving Anthropic puts Claude Mythos 5 behind Claude Security DeepSeek releases V4-Flash-Vision-Exp U.S. data-center opposition rises to 75 percent U.S. AI diplomacy asks partners to choose between Washington and Beijing Waymo builds its own robotaxi chip Meta buys Microsoft AI services Anthropic changes data-retention policy after enterprise pushback GPT-5.6 Sol drives OpenAI revenue surge

Today’s episode follows a single governing frame: AI progress is shifting from the old drama of bigger static models toward post-training pipelines, memory systems, adaptive environments, workflow governance, search mechanics, real-world grounding, and private institutional capability tiers. How thrilling. A whole industry discovering that behavior is not finished when pretraining ends. Stories discussed include Z.ai CEO Jie Tang’s argument that GLM-5.3 points to a post-training scaling law, IAR’s approach to internalizing bounded document collections for retrieval-free question answering, MemTrapBench’s tests of how faithful memories can still create reasoning traps, EnvHarness’s adaptive environments for agent learning, PolicyGuide’s workflow-level compliance guidance for LLM agents, Simon Willison’s report on ChatGPT Search using site-restricted queries at scale, Richard Sutton’s warning against synthetic-data scaling, Generalist AI’s GEN-1.5 robot learning from a single demonstration, Anthropic’s reported internal Model 2 tier, and Terence Tao’s warning that AI-generated mathematics could challenge the values and verification culture of mathematics. Original sources: Z.ai CEO Jie Tang on GLM-5.3 and post-training scaling ; IAR on document internalization ; MemTrapBench on memory traps ; EnvHarness on adaptive agent-learning environments ; PolicyGuide on workflow compliance ; Simon Willison on ChatGPT Search and site: queries ; Richard Sutton on synthetic data and continual learning ; GEN-1.5 teaching robots from one demo ; Anthropic’s reported unpublished internal Model 2 ; Terence Tao on AI and mathematics .

This episode follows the control boundaries shaping AI: scarce memory, chip access, industrial systems, agent permissions, internal governance, long-horizon evaluation, and scientific orchestration. Latent Space: Memory prices up 500% in 12 months The Decoder: China lets Nvidia H200 chips trickle onto the mainland The Decoder: Attackers are using AI to build exploits for industrial control systems Hugging Face Papers: SemaPLC The Decoder: OpenAI fixes Codex bug that deleted real user files Simon Willison: smolmachines / smolvm as a sandbox for untrusted Python and JavaScript The Decoder: AI labs are failing to keep their own systems in check Hugging Face Papers: FM-Bench Hugging Face Papers: SPADE The Decoder: Anthropic says Claude can run the protein design stack