OpenAI chief scientist: losses below 0.8 unlocked emergent cybersecurity and math capabilities
kimmonismus · x · 2026-11-23
kimmonismus highlights a quote from OpenAI chief scientist Jakub Pachocki: once pretraining losses dropped below 0.8, transformative capabilities emerged in cybersecurity and mathematics. Pachocki stresses it was not at all clear a priori that capabilities would appear at those precise perplexity thresholds — they are best understood as emergent latent capabilities from progress on this single metric.
It's the official footnote to the loss-curve reading: capability jumps map to perplexity thresholds, and near the floor even small loss reductions imply outsized intelligence gains.
More from AGI Musings
- Harari to Suleyman: 'If they are really gods, they will not remain your slaves' — PeterBowdenLive · 2026-11-23
- Two-person lab GSI claims 250+ breakthroughs across 8 fields with custom AI-built solvers, not brute-force compute — udmrzn · 2026-11-23
- iamtrask: Great writing is constrained by IP access, proofs only by compute — iamtrask · 2026-11-23
- Daniel Litt: the Lean-formalized Riemann proof is likely correct, no human understands it yet — littmath · 2026-11-23
- Physicist Anthony Aguirre publicly challenges AGI lab staff: you're building superintelligence that makes humans cognitively obsolete — michael_nielsen · 2026-11-23
- Healthcare AI advocate to doctors: AI changes the workflow, not the mission of medicine — HealthcareAIGuy · 2026-11-23