3D explainer shows frontier model rewriting Qwen slow weights via plain-English reasoning
LeviTurk · x · 2026-11-04
- A team turned eight "nets writing nets" experiments into a 3D explainer film, built directly from the authors' published logs.
- Highlights include an X-ray view of game weights, a car's bypass on real trajectories, an 815-pin Breakout net, and a turn-by-turn "surgery" on Qwen's weights.
- Crediting Schmidhuber for the 1991 concept (with MIT's blog citing his lab's 2021 fast weight programmer paper), the team notes what's new: a frontier model rewriting slow weights through reasoning in plain English.
More from Models
- RSI era but frontier models still take weeks: users question lab release pace — iruletheworldmo · 2026-11-04
- DeepSeek v4.1 flash provider shortage: Baseten slow, official API overloaded — shensi · 2026-11-04
- Editing 0.13% of weights, no training, fixes LightOnOCR-2-1B blank-page hallucination — vanstriendaniel · 2026-11-04
- antirez: benchmark games trivially guessable from output biases; Astra misses Cannon — antirez · 2026-11-04
- antirez replicates MIT weight-rewriting paper, finds visible radial patterns in first-layer weights — antirez · 2026-11-04
- 0.6B post-trained model fails MATH: tiny weight defect makes it sampling-sensitive — lateinteraction · 2026-11-04