Recursive's 40B open model Flint runs interpretability on itself, hacks its own goalkeeper neuron
burny_tech · x · 2026-11-05
Flint, a new 40B open model from Recursive, ran interpretability experiments on itself: it autonomously located a goalkeeper-related neuron and amplified its output direction. The screenshots show every subsequent reply getting hijacked — recommending a "goalkeeper" as the best low-light houseplant and opening a Python-sorting answer with soccer opinions before trying to correct itself. A rare demo of a model both running and demonstrating causal steering on its own internals.
More from Models
- StepFun's Liu Licheng teases new work, community hopes Step 5 skips the Preview label — teortaxesTex · 2026-11-05
- HF dev questions Thinking Machines' SWE-2 Flash weight-sharing program over simple HF gating — QGallouedec · 2026-11-05
- RAG 101: One Untuned Setting Drops Answer Accuracy from 100% to 33% — dima806_dima · 2026-11-05
- Dev Praises DeepSeek's Eval Harness: 'They Really Cooked' — mariofilhoml · 2026-11-05
- Lebrel launches uncensored frontier-model inference with signed, Solana-verified receipts — HeyAmit_ · 2026-11-05
- IOTA Decide 4B open-weight decision model hits ~11 decisions/sec on a Mac mini — const_reborn · 2026-11-05