Back to News

Researchers: LLMs can natively read and write neural network weights; GPT-6 Astra edits Qwen3.5-0.8B to boost math

#llm#neural-network-weights#model-editing#gpt-6-astra

Researchers report that large language models can natively read and write neural network weights, demonstrating this by having GPT-6 Astra edit the weights of Qwen3.5-0.8B to improve its math performance. The finding suggests LLMs may be capable of directly manipulating model parameters, opening new possibilities for model editing and optimization.

Coverage timeline

  1. Techmeme
    Researchers say LLMs can be good at natively reading and writing neural network weights; GPT-6 Astra edited Qwen3.5-0.8B's weights to make it better at math (Rikiya Takehi)

    Rikiya Takehi : Researchers say LLMs can be good at natively reading and writing neural network weights; GPT-6 Astra edited Qwen3.5-0.8B's weights to make it better at math — TL;DR … If you were asked to handcraft a neural network that plays Go, what would you do? If you were handed raw neural network weights …

Related stories

  • Goodfire researcher argues neural networks are largely interpretable, most neurons understandable

    Nick Cammarata of Goodfire published an essay arguing that neural networks are far more interpretable than previously believed, with most neurons in models up to trillions of parameters understandable. He attributes polysemanticity in early models to fixable issues rather than superposition, and suggests mechanistic interpretability may be solved soon.

    3 reports · 3 sources

  • UCSB and Microsoft Propose ScholarEvolve for AI Agents to Read Papers and Self-Modify Code

    Researchers from UC Santa Barbara and Microsoft introduced ScholarEvolve, a method that automates the research process of reading literature, implementing algorithms, and evaluating results to enable AI agents to evolve their harness code without modifying underlying model parameters. The approach addresses the limitation of prior coding-agent-based harness optimization, which relied on the agent's existing knowledge and observed failures, by allowing agents to draw on the latest academic findings. The paper and code are publicly available.

    2 reports · 1 sources

  • SenseTime-led study: GPT-6-Astra scores near-perfect on benchmarks but fails real-robot tasks

    A research report by SenseTime, Daxia Robotics, Ropedia, NTU Singapore, and the University of Tokyo analyzes GPT-6-Astra's embodied intelligence across 12 simulation benchmarks and 8 real-robot platforms. The model achieves near-perfect scores on traditional benchmarks (e.g., 100% on LIBERO-Long++, 99.3% on LIBERO-Pro) but fails on real-robot tasks, such as taking over 4 hours and $55 to place a pen in a holder, and failing 11 insertion attempts. The report concludes GPT-6-Astra has largely crossed the embodied intelligence threshold but remains far from practical deployment.

    1 reports · 1 sources

  • Startups use machine unlearning on Chinese open-weight models to remove censorship

    Rest of World reports that startups like Hirundo and Lazarus are applying machine unlearning and weight edits to Chinese open-weight models to remove censorship and harden security for Western companies. This trend reflects growing Western business concerns about censorship and security in Chinese models.

    1 reports · 1 sources

News updates every hour. Subscribe in your RSS reader to get new stories automatically.

Subscribe