← sources

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)

source: techmemepublished 2026-11-02 20:00ingested 2026-11-03 02:45

retained text

Techmeme permalink: https://www.techmeme.com/261102/p36#a261102p36

<a href="https://rikiyat.github.io/blog/weight-space-programming/"><img align="RIGHT" border="0" hspace="4" src="http://www.techmeme.com/261102/i36.jpg" vspace="4" /></a>
<p><a href="https://www.techmeme.com/261102/p36#a261102p36" title="Techmeme permalink"><img height="12" src="http://www.techmeme.com/img/pml.png" style="border: none; padding: 0; margin: 0;" width="11" /></a> <a href="https://rikiyat.github.io/">Rikiya Takehi</a>:<br />
<span style="font-size: 1.3em;"><b><a href="https://rikiyat.github.io/blog/weight-space-programming/">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</a></b></span>&nbsp; &mdash;&nbsp; TL;DR &hellip; If you were asked to handcraft a neural network that plays Go, what would you do?&nbsp; If you were handed raw neural network weights &hellip; </p>

document fetched 2026-11-03 02:32 · canonical https://techmeme.com/261102/p36