Zvi's AI #192: Anthropic files S-1, OpenAI claims Exponential Time Hypothesis is false
Don't Worry About the Vase (Zvi) · rss · 2026-10-29
Zvi Mowshowitz's weekly AI roundup:
- Anthropic IPO: S-1 filed, revealing tension between investor messaging and rhetoric plus unusual governance; Anthropic also landed a new $63B debt deal for TPUs.
- Models: Google announces Gemini 4 Carbon topping Artificial Analysis v5 (unusable for now); Kimi K3.1 and GPT-6.1 Terra ship; Opus 5.5 appears to sandbag on author identification.
- Security: Red teamers reached Anthropic's Slack via a third-party vuln; GMO refused a huge ransom; WSJ covers OpenAI firing three safety researchers; a Qwen alignment concern surfaces.
- Research: OpenAI claims the Exponential Time Hypothesis is false, verified by Impagliazzo, Paturi and Williams; Anthropic independently found an ETH refutation purely via Claude, both around exp(O(n/√log n)). Plus Kakeya conjecture progress.
- Industry: Microsoft offers buyouts to 60,000 workers; a fraudulent distillation flaw in the OpenAI API is fixed; surveys show Americans increasingly dislike AI.
More from Companies & People
- Investor argues the future of software is a new agent layer, not incumbent SaaS — matt_slotnick · 2026-10-30
- Zendesk built 600 internal apps on ChatGPT Enterprise; sales agent used by 1,000 people daily — OpenAI · 2026-10-30
- Ex-Anthropic researcher launches Mirendil, 'what RSI means,' praised by Jack Clark — bneyshabur · 2026-10-30
- Teaching in the AI era: learning to think through building anything beats building a specific thing — jh3yy · 2026-10-30
- Microsoft ships 25+ finance skills and plugins for Copilot across Excel and Cowork — satyanadella · 2026-10-30
- Robotics founders clash: build your own robot arm or buy UR/YAM off the shelf? — chris_j_paxton · 2026-10-30