Company
Company behind AI Scientist-v2, which produced a fully AI-generated paper accepted at an ICLR workshop.
Sakana AI and UC Berkeley's MASS method reached a 69.9% win rate against its Qwen3.6-27B base model after two rounds of self-training. The same model ranks its own runs, and its evaluator sometimes passes flawed steps into the next training set.
Reality
MIT and Sakana AI's SIFT ran a coding agent's full self-improvement search on roughly $34 of API calls, about a tenth of the Darwin Godel Machine's resources. The saving comes from a language-model judge screening patches, so it holds only while that judge picks correctly.
Reality
Tencent's Hunyuan Speech team put a swappable agent model behind a conversation model that re-decides every second whether to talk, and reports best-in-test timing on Full-Duplex-Bench v3 with a task-accuracy shortfall it describes only as slight.
Reality
In simulation the cubes kept their answer with 15 percent of them deleted and rebuilt the gaps, and in hardware 197 blocks named their object every time, which is a smaller and cleaner claim than fault-tolerant computing.
Reality
A theoretical paper assumes error-free, near-free language models and still finds that in two of three deployment scenarios, researchers respond by doing more work with less care.
Reality