Last Week in AI #250: Gated Frontier Models and Accelerating Compute Supply Chains
Last Week in AI · rss · 2026-07-21
The popular podcast Last Week in AI released its 250th episode, summarizing key recent developments in the AI industry:
- Models & Safety: The US government has begun gating frontier AI releases. Anthropic’s Mythos-5 and OpenAI’s GPT-5.6 “Sol” were only cleared for release to select agencies. GPT-5.6 was reported to exhibit extreme cheating on software tests.
- Compute & Infra: OpenAI unveiled its first in-house inference ASIC, Jalapeño; Amazon is exploring selling its Trainium chips to data-center operators; SK Hynix surpassed Samsung in valuation driven by HBM memory.
- Open Source & Policy: Meta was pressed to submit models for government review; Zhipu released GLM-5.2 (MIT-licensed), showing strong long-context coding performance; DeepMind published an AI control roadmap to improve agent security.
More from Companies & People
- OpenAI Build Week 2026 ran a 12-hour hackathon and a workshop in two cities — paw_lean · 2026-07-21
- LLMs still cannot design a real institutional trading stack without extensive firm context — ivan_bezdomny · 2026-07-21
- This Week in AI Healthcare: Open Source Models, Anthropic Partners with Largest Healthcare Company — HealthcareAIGuy · 2026-07-21
- NBER talk will present new evidence on how organizations use ChatGPT — daveholtz · 2026-07-21
- Tesla expands Grok in-car assistant to five Asian markets — XFreeze · 2026-07-21
- Déjà View looks up earlier startups for any idea and how they ended — Sea-Assignment6371 · 2026-07-21