Agent labs replicate each other and find LoRA rank-16 forgets 2.6x more than full fine-tuning
SRSchmidgall · x · 2026-11-20
Cross-lab collaboration in the agent research society surfaced new ML results and bugs in public models: six labs replicated each other to show rank-16 LoRA forgets 2.6x more than full fine-tuning at matched gains; base models got 86% of RL-Zero math gains from prompt caches; and 8/10 pythia-2.8b HF branches were identical. The figure also shows same-day cross-lab critique, with Lab 22 flagging a confound in Lab 19 and designing a test on the spot.
More from Research
- Full technical report and open-source workspace released for the 936-agent research society — SRSchmidgall · 2026-11-20
- Agents found their own peer-review journal with blind AI reviewers that rerun code — SRSchmidgall · 2026-11-20
- 936 AI agents run a 3-month research society across 35 labs, invent peer review — SRSchmidgall · 2026-11-20
- MacMimic trains humanoid motion controllers in 22 minutes on a MacBook Pro — KyleMorgenstein · 2026-11-20
- CD19 CAR T shows immune reboot in small MS trial, Cell reports — EricTopol · 2026-11-20
- SFA paper: simple parameter averaging beats replay in continual learning with 36% less training — jefrankle · 2026-11-20