trash research. guys please read it for fucks sake
Will Depue calls the @_doubleAI_ LLM re-architecture research trash
AIWill Depue, who is listed as an OpenAI account owner, posts that the @_doubleAI_ research is "trash" and urges readers to read it. The quoted thread says coding agents spliced code into networks during training, raising nanochat loss below baseline and algorithmic reasoning accuracy from 55% to 84%.
will depueVerified on X
@willdepue
1/ What if an LLM could not just edit a model's weights, but re-architect it as it trains? We tried just that. The resulting architectures exit the known territory of transformers: the coding agents spliced code directly into the network. On nanochat, the loss fell below ordinary training after each new code layer was introduced. On algorithmic reasoning, accuracy rose from 55% to 84%. We even let DeepSeek-V4.1-Flash act as the agent modifying itself, raising its AIME and HMMT 2026 score from 64% to 71%. To do all this, the agents built a set of tools we’ve never seen before, which we call mechanistic intervention! Blog: https://doubleai.s.gy/llm-redesign
Source details
will depue · x.comPublished · added here
Original title: trash research. guys please read it for fucks sake