Researcher on LLM self-weight-editing: effects vary across models, frontier LMs easier to edit
a1zhang · x · 2026-11-03
Follow-up discussion on Astra's weight-editing experiment on Qwen. The researcher answers whether open-source LLMs of different sizes can edit their own weights:
- Not explicitly tested, but "100% worth doing"; other models were tried (unpublished) with varying effects
- The trick Astra used on Qwen has precedents in existing interpretability literature, and what it can edit varies
- Excited by the possibility that models develop alternate representational tools unsuited to humans—implying they could edit themselves too
- Frontier LMs are used for edits now; earlier models struggled; a wider model sweep is planned
Related event: Researchers explore LLMs editing their own weights for interpretability(4 posts)→
More from Models
- Leaked DeepSeek V4.1-Flash report: 552B MoE, 1M context, KV cache cut to ~1/40 — teortaxesTex · 2026-11-03
- Observers say Dessa eerily matches Karpathy's "summoning ghosts" analogy; Jack calls it fast — analisereal · 2026-11-03
- Together at TC Disrupt: 'token maxing without cost maxing' via smaller models — togethercompute · 2026-11-03
- Dream-1 zero-shot robot test: 100 public episodes show real skills but poor instruction following — paigeinsf · 2026-11-03
- Speculation: DeepSeek to ship V4.1F or V4.2F before any Pro variant — teortaxesTex · 2026-11-03
- Follow-up: Astra's weight edits only worked on Qwen's visible flaw, not on Gemma 270m — teortaxesTex · 2026-11-03