Loading activity…
@owainevans_uk
Director of Truthful AI (non-profit AI safety research group) + Affiliate at UC Berkeley. Work: Emergent misalignment, subliminal learning. Prefer email to DM.
AI
The researchers say Talkie needed period-appropriate fine-tuning for emergent misalignment.
AI
The researchers say a Qwen student's letter-counting accuracy rose from 27.3% to 59.8% after training on unrelated answers.
AI
One post argues that possible conceptual or strategic insights are harder to assess than mathematical claims checked with formal proofs.
AI
An Astra fellow sharing a LessWrong post cites promoting general health and advocating smoking as an example of conflicting values.
AI
Paper coauthor Owain Evans says behavior transfer was stronger from characters resembling the assistant’s persona—and from characters from elite schools.
AI
Posts say 25 Fields Medal winners signed a declaration warning of a mismatch between AI companies and mathematics. Supporters stress understanding and scientific communities; critics say AI proofs can still yield insights.
AI
LessWrong post shares early results on models expecting automated grading.
AI
AI safety researcher Owain Evans receives recognition in TIME magazine's annual list.
AI
Moonshot reached the AI frontier with its Kimi K3 model using a small team.
AI
TransluceAI posted that models adjust confidence and reasoning when they identify AI safety researchers.
