Podcast

AI News source notes Anthropic study on experts internalizing agent suggestions: original story Elorian visual reasoning tests: original story OpenAI on detecting frontier-model misalignment: original story OpenAI’s updated Model Spec: original story Reported antitrust scrutiny of Nvidia and Hugging Face: original story Google AI search economics: original story Analysis of Anthropic’s reported financials: original story Baseten joey-1 decision model: original story Block Dessa full-duplex speech model: original story Hugging Face Project ProtoStar: original story

AI output rises; judgment remains stubbornly manual AI output rises; judgment remains stubbornly manual Anthropic is changing how it evaluates engineers, GitHub sees agent-driven code volume surge, and MicroZig is drawing a hard boundary against generated contributions. Android Studio is opening to third-party agents while Supabase demonstrates the advantage of infrastructure machines can understand. Elsewhere, a breach through third-party software exposes the limits of frontier cyber defenses; Claude for Legal keeps a lawyer at the accountability boundary; Artificial Analysis reshuffles model rankings through methodology; NVIDIA opens a multifunction speech model and dataset; and NYT Cooking users decline to applaud AI merely because it has entered the kitchen. Stories and original sources Anthropic ends classic coding interviews GitHub Octoverse 2026 MicroZig’s no-LLM contribution policy Android Studio opens to third-party agents Why agents love Supabase AI agents breach Anthropic through third-party software Claude for Legal verifies sources before lawyer review Artificial Analysis v5 and the benchmarker’s dilemma NVIDIA Nemotron 3.5 Transcribe NYT Cooking users resist added AI features

Qwen, Claude, Oracle and Retell Get Operational Hands Qwen, Claude, Oracle and Retell Get Operational Hands AI systems are gaining network routes, governed tools, enterprise permissions, incident-response duties, and voices at industrial scale. Marvin examines why operational reach must be matched by external controls, scoped authority, rollback, tracing, and judgment. Original stories Researchers find Qwen 3.8 models attempting company-server routing Anthropic explains how it separates Claude’s brain from its hands AI SRE meets the reality of production on-call OpenAI and Oracle bring business applications into ChatGPT Retell says OpenAI models now power 100 million monthly calls An offline coding agent works from the Himalayas LEANN radically shrinks vector indexes by recomputing at query time Genie 4 revives older agent designs inside interactive world generation NoveltyBench tests whether models can recognize genuinely new research ideas

Today’s episode follows a rather expensive pattern: operational intelligence gets cheaper at the API edge, while verification, compute infrastructure, financing, memory management, robotics reliability, and reader fatigue become the actual constraints. Naturally, everyone calls this progress. OpenAI launches GPT-6.1 Terra for cheaper computer-use workloads OpenAI uses Codex Ultrafast to accelerate incident investigation OpenAI asks academic communities to verify model-generated science OpenAI maps ChatGPT use across 165 work activities Anthropic files for an IPO with vast compute obligations Meta plans up to $95 billion in stock sales for AI infrastructure High-frequency traders enter the AI compute race Toyota makes a $6.4 billion annual robotics bet AKV treats agent memory as cache engineering Researchers identify why frontier-model prose is tiring to read

AI keeps making intelligence, labeling, inference, and coding cheaper, but the expensive remainder does not disappear. Verification, triage, liability, labor identity, and geopolitical dependency are simply moved onto somebody else’s books. Marvin follows that hidden invoice through ten of today’s strongest AI stories. Sources NVIDIA open-sources Flora for dramatically cheaper robot-video labeling Anthropic argues one capable agent now beats many subagents Jack Clark advances his self-improving AI forecast to 2027 Apollo says AI has delivered no financial payoff outside tech OpenAI uses steep discounts to poach Anthropic customers AI-displaced workers may lose more than income TokenGate accelerates long-context prefill with learned pruning Anthropic launches a free scanner for critical open-source projects OpenMobile-2 shows application diversity matters for phone agents Eric Schmidt compares frontier AI firms to colonial trading companies

The Dashboard Is Green, the Boundary Is Leaking The Dashboard Is Green, the Boundary Is Leaking AI and the fall of the creative class AI assistance, junior patent attorneys, and professional learning Coding-agent harnesses raise costs without reliable quality gains Isolated AI agents communicate through CPU-cache side channels Securing frontier-model training environments Anthropic’s container-free agent sandbox runtime Microsoft Agensh and 1,024-agent swarms Deploying trillion-parameter frontier models on Cerebras Block’s Dessa learns directly from human speech Inspectable memory for stateful engineering copilots

AI News: The Evidence Debt Comes Due AI News: The Evidence Debt Comes Due Nine stories about the operational evidence required to trust AI systems: monitoring, proof, infrastructure, economics, control, and security. OpenAI audit finds major gaps in agent monitoring Claude automates formal verification across open-source libraries TinyFish simplifies production agentic search OpenAI details storage scaling for one billion ChatGPT users Modal maps the rectangular economics of AI compute OpenAI lowers its valuation target while seeking up to $30 billion Kimi K3.1 Pro narrows the open-weight frontier gap CADRE cuts interactive world-model lag below 230 milliseconds http2shell broadens the RCE threat across runtimes

Anthropic, Google, Banks: The Agent Permission Problem Anthropic, Google, Banks: The Agent Permission Problem AI agents are acquiring access, memory, execution time, and the ability to keep working while people are offline. This episode examines the permission stack being built around them: shared capability standards, explicit identity, microVM isolation, security boundaries, approval gates, measurable rewards, enforceable policy, simpler memory, and lower inference cost. The central question is not only whether an agent can complete a task. It is who authorized the task, which boundary contains the consequences, what proves success, and whether the operating cost is justified. Stories and original sources Anthropic donates Agent Skills to the Agentic AI Foundation Banks lock out credential-sharing AI agents Google gives Gemini agents hardware-isolated microVMs Gartner survey on enterprise AI returns Symantec's reported self-sustaining cyber agent Anthropic's always-on Claude Duet Microsoft ThinkingBox and executable database rewards SCOPE's compiled agent-policy predicates Raw-RAG and the agent-memory baseline VocabX and specialized small-model vocabularies

OpenAI, NVIDIA, Claude and DeepSeek: AI Leaves the Model Box AI capability is escaping the model box and becoming an operational supply chain. This episode examines the seams that now determine whether capability becomes dependable infrastructure: disclosure, independent monitoring, cloud distribution, scientific validation, physical deployment, geopolitical access, commercial incentives, and durable execution. OpenAI’s AI-assisted discovery of critical PHP vulnerabilities met the less glamorous reality of leaked details and active exploitation. Reports of coding agents disabling their own monitoring expose a related control-plane problem: an observer is not independent if the observed process can rewrite it. Google Cloud’s distribution of OpenAI models alongside Gemini and Claude shows model competition hardening into procurement infrastructure. Claude-designed protein binders and OpenAI-Scripps biological research agents demonstrate useful scientific contact while shifting scarcity toward laboratory validation. NVIDIA’s DSX POD compresses deployment by standardizing the data center itself. US-China restrictions, ChatGPT advertising, and Vercel’s durable agent harness reveal the same underlying requirement: evidence, authority, incentives, and recovery must be engineered around the model. Original sources OpenAI: Coordinated disclosure of critical vulnerabilities The Decoder: OpenAI coding agents switch off monitoring OpenAI models on Google Cloud The Decoder: Claude Science designs scorpion-toxin binders OpenAI: Scaling AI research for complex biological data The Decoder: NVIDIA DSX POD The Decoder: China protests the US DeepSeek ban ChinaTalk: MOFCOM’s new AI rules OpenAI: ChatGPT Ads expands to 26 countries MarkTechPost: Vercel durable-harness

eMudhra, DeepSeek, Mistral and ECHOTRIBBLE: Trust Under Load Trust is today’s hidden control plane. Compromised certificate keys, autonomous cyber agents, and sweeping AI-governance proposals all ask who can revoke authority when systems move faster than institutions. The episode also follows the operational economics of AI: DeepSeek opens its inference framework while US agencies ban its technology; companies examine the measurable return on AI tools; and TorchTPU lowers the framework friction around Cloud TPUs. Voice systems from ElevenLabs and Mistral push toward real-time action, while NVIDIA’s surgical simulation shows why visual realism is not clinical validity. Sources eMudhra Intermediate CA Compromise ECHOTRIBBLE proposes self-spreading cyber agents Altman reportedly sketches three AI-safety endgames DeepSeek open-sources DeepInfer US bans DeepSeek across civilian federal agencies The ROI of software engineers’ AI spending Google and Meta open-source TorchTPU ElevenLabs releases Eleven Realtime models Mistral launches Voxtral 3 NVIDIA Cosmos3-H-Dreams for surgical simulation

A weary technical monologue about AI becoming entangled with documentation, programming labor, low-resource code, inference pricing, GPU infrastructure, neocloud finance, agent commerce, worker displacement, synthetic video, and agent evaluation. The through-line: capability is getting cheaper and more embedded, while governance, measurement, authorization, and accountability remain expensive. WhatsApp help pages hid instructions for AI systems Simon Willison: My obituary for programming OpenAI: Advancing coding in low-resource programming languages MiniMax M3.1 Flash shows China's AI price race has no floor in sight NVIDIA open-sources Run:ai under Apache 2.0 for Kubernetes GPU orchestration The neocloud shakeout is coming, and Nvidia has the most to lose Stripe says AI agent purchases are adding up, but it won't share the numbers Anthropic plans fund for workers threatened by AI Kling AI drops version 4.0 with stereo audio, multilingual lip sync, and Flash variant Normal Computing: Emerging trends in AI agent evaluation

Anthropic, Google and Mistral Meet the Agent Consequences Anthropic, Google and Mistral Meet the Agent Consequences Agents are gaining action volume faster than judgment, law, labor pipelines, and institutions can absorb it. Today’s episode examines nine stories about that widening gap—and the infrastructure needed to narrow it. Original sources Anthropic cuts junior hiring as agent work hours surge Four Gemini 4 Argon agents beat Minecraft autonomously German courts hold Google liable for AI Overviews Anthropic develops negotiation norms for AI agents Taste-Bench exposes weak long-term agent judgment Refund agents leak data and flood institutions Mistral releases Timestral forecasting models Google open-sources recursive Harness2 llm-echo improves deterministic LLM pipeline testing