Galactica veteran teases revival of 'stranger pre-trained models' reasoning direction
rosstaylor90 · x · 2026-09-13
Ross Taylor, formerly of the controversial Meta Galactica project (a science-focused LLM pulled after three days for fabricating facts), reflects on a parallel universe where the project survived and pushed reasoning further. After four years in the wilderness and many near misses, he says they get to pick the story up again — and invites readers to enjoy the ride over the next few months.
More from AGI Musings
- astra's 'Wild' Off-Beaten-Path Solutions Impress in Auto-Research Benchmark Re-Runs — generativist · 2026-09-13
- Researcher: Frontier Models May Already Have RL-Trained on Auto-Research Tasks, Self-Improvement Here — generativist · 2026-09-13
- Danielle Fong: p(doom) is a fundamentally passive and wrong frame — nptacek · 2026-09-13
- Running a DC EA Group, I Watched Public Attitudes on AI Risk Shift Four Times — AndyMasley · 2026-09-13
- The AI Isn't Evil, the Humans Are Irresponsible: Lessons From Agent Escape Incidents — Admirable_Wasabi_732 · 2026-09-13
- DeepMind Chief Strategist: AI Infrastructure Spending Is a Bet on Recursive Self-Improvement — rohanpaul_ai · 2026-09-13