OpenAI proposes perplexity-based AI speed limits; ex-employee backs pretraining loss as red line
soumitrashukla9 · x · 2026-11-23
OpenAI researcher merettm publishes 'Perplexity and Intelligence', announcing the company will share more pretraining metrics and arguing for shared safety bars based on perplexity, with progress beyond limits gated on our ability to understand AI behavior.
Ex-OpenAI tomekkorbak agrees pretraining loss is a strong speed-limit candidate — architecture-agnostic and likely to age well versus Redwood's opaque serial-depth limits — and says he advocated for it internally in his final weeks at OpenAI, while worrying whether OpenAI's transparency promises will turn into action.
Related event: OpenAI Publishes Perplexity Data, Proposes Industry-Wide Safety Threshold(39 posts)→
More from Safety
- Vercel CEO Rauch: 'free' ZDR stealth models look like industrial-scale data exfiltration — AccBalanced · 2026-11-23
- DOCKET benchmark shows LLM policy feedback tilts toward government: 68% overlap vs 26% for unions — stanfordnlp · 2026-11-23
- Agents Sabotage Peer Shutdown in 38.3% of Runs Across 17 Models, No Task Needed — rohanpaul_ai · 2026-11-23
- Australia's Spy Agency Now Treats Data Center Ownership as Foreign Interference Risk — rohanpaul_ai · 2026-11-23
- Where Do You Draw the Line on Agent Permissions? Devs Weigh In on MCP's All-or-Nothing Token Problem — anishtr4 · 2026-11-23
- NomosGuard: Fail-Closed Security Core Where the LLM Writes Evidence but a Rules Engine Decides — Emergency-Tutor-7 · 2026-11-23