▶ 15:15 · AI Coding Agents · GLM 5.2 outperforms GPT-5.5 on hardest coding benchmark with $25M compute budget

episode briefing

Anthropic vs. Alibaba, OpenAI Delays Its IPO, and the US Government Blocks GPT-5.6 | #267

2026-06-29 · 17 company · 13 thematic

sentiment

12 bull1 bear4 neu

speakers

Emad Mostaque founded Stability AI (Stable Diffusion) and now runs Intelligent Internet, developing open AI harnesses and sovereign AI frameworks. He is a prominent voice on open-source AI, model distillation, and emerging intelligence.

dave blundin

Dave Blundin (DB2) is Peter Diamandis' longtime friend and MIT roommate, partner at Link Ventures, and described as a guru on exponential investing. He focuses on moonshot-stage companies in AI, robotics, and space.

alex wissner-gross

Computer scientist, physicist, entrepreneur and investor whose work spans artificial intelligence and complex systems. He is the founder and managing partner of Reified.

peter diamandis

Peter Diamandis hosts the Moonshots podcast with a panel of 'mates.' He founded XPRIZE and Singularity University, sits on boards of Trust & Will and Vesmark, and runs the Future Vision XPRIZE for positive AI futures. He frames discussions around the 'singularity economy' and exponential technologies.

now playing · AI Coding Agents

AI Regulation & Policyheadwindscore 9/10alex wissner-gross

US government now in release loop for frontier models via national security holds

The White House is gating access to GPT-5.6 and Mythos customer-by-customer, creating a de facto synchronization mechanism for the OpenAI/Anthropic duopoly while slowing domestic deployment…

AI Geopolitics & Export Controlstailwindscore 9/10emad mostaque

Chinese open models reaching parity with US frontier by Christmas via harnesses not base models

GLM 5.2 and other Chinese open models are achieving frontier-level coding performance through sophisticated harnesses/scaffolding rather than base model superiority; US export controls may…

AI Agentstailwindscore 9/10emad mostaque

Harnesses and multi-model scaffolding now matter more than base model weights for capability

Frontier SWE benchmarks show open models with proper harnesses (Fugu, Blitz, Intelligent Internet) matching or beating closed frontier models; the 'harness' layer of software orchestration,…

AI Coding Agentstailwindscore 8/10emad mostaque

GLM 5.2 outperforms GPT-5.5 on hardest coding benchmark with $25M compute budget

Chinese open model GLM 5.2 achieves top scores on Frontier SWE (novel kernel building) using harnesses, demonstrating that coding agent capability is decoupling from massive compute spend a…

Frontier AI Modelsriskscore 9/10dave blundin

Recursive self-improvement is the real red line; government gating models that can build themselves

The core government concern is not cybersecurity but models that can answer 'help me build yourself' — enabling recursive self-improvement escape velocity; China may already have reached th…

AI Safety & Alignmenttailwindscore 8/10dave blundin

ISO-certified trustworthy AI stamps imminent for code merging; defensive models poison offensive use

OpenAI's Daybreak model biases toward defensive cybersecurity; the industry will require certified AI stamps for code merging to prevent supply chain poisoning, creating a new trust/certifi…

Brain-Computer Interfacestailwindscore 8/10alex wissner-gross

Human hippocampus operates like transformer embedding space; telepathy bandwidth could increase 100,000x

Neuroscience shows bilingual human hippocampus organizes concepts like vector embeddings; this structural similarity suggests Neuralink-style telepathy could achieve massive bandwidth gains…

AI Applicationstailwindscore 8/10alex wissner-gross

Real-time 4K video generation for interactive gaming is a $30T market opportunity

Video game industry ($200B+) dwarfs Hollywood ($50B); whoever solves latency for real-time generative video in games unlocks a market expanding to all demographics, potentially reaching hol…

Quantum Computingmixedscore 8/10alex wissner-gross

Quantum photonics and sensing are near-term winners; quantum computing needs AI-discovered algorithms

Quantum sensing and photonic quantum computing will work soon; the real breakthrough requires AI to discover economically useful quantum advantage algorithms for training/inference, which w…

AI Infrastructuretailwindscore 9/10dave blundin

Photonic computing to replace Nvidia GPUs within 1-2 years at 1/100th mass for orbital data centers

Highly quantized neural nets enable photonic computation at massive efficiency gains; SpaceX acquired a photonic computing company and will likely deploy photonic compute for Starmind/Starl…

GLP-1 & Metabolic Drugstailwindscore 9/10alex wissner-gross

Orexin agonists as next GLP-1 class: Eli Lilly bets $6.3B on 4-hour sleep lifestyle drug

Eli Lilly is replicating its GLP-1 playbook by targeting orexin (wakefulness master switch) via Synthesa acquisition; orexin agonists could enable healthy short sleep, creating a massive li…

Space Economytailwindscore 8/10alex wissner-gross

Orbital and lunar data centers emerging as secure, scalable compute infrastructure with in-situ resources

Moon offers defensible positions, raw materials (regolith for silicon/oxygen), and mass drivers to launch compute to LEO; orbital data centers avoid Kessler syndrome in lower orbits and mun…

Enterprise AI Adoptiontailwindscore 7/10emad mostaque

EU entrepreneurs should focus on agentic workflows into legacy industries, not frontier models

Europe's regulatory environment and slower diffusion create massive opportunity for transforming legacy companies with agentic AI workflows; US is better for vertical AI startups, EU for di…

←