Cohere researcher: labs should share validation loss and idea throughput, not efficiency gains
eliebakouch · x · 2026-11-06
Cohere's Elie Bakouch argues there's a spectrum between hoarding confidential info and sharing nothing: pre-training efficiency gains over previous architectures don't help competitors even if disclosed, so labs could share things like monorepo validation loss (as OpenAI used to), SFT->RL SWE-bench results, or hinted scaling laws. He also suggests quantifying METR interview claims — e.g. ideas generated per researcher per week with Claude + scaffold.
More from Research
- Google open-sources Agile Deliberation, an AI framework that turns fuzzy classification rules into explicit ones — aigclink · 2026-11-06
- Systematic review: GenAI fosters or hinders critical thinking depending on how it's used — _akpiper · 2026-11-06
- Developer Validates AME-2 on Humanoids: 70cm Foot-Only Climbing, 1.2m Gap Traversal in Sim — ChongZzZhang · 2026-11-06
- Citizen science labeling works better when small groups deliberate, not annotate alone — _akpiper · 2026-11-06
- Wayve releases Learned Safety strategy after a decade of end-to-end AV research — alexgkendall · 2026-11-06
- Stranger uses open-source AI to model ALK resistance and design backup cancer drug for free — anshulkundaje · 2026-11-06