Microsoft's iWSP: Claude writes transformer weights directly, 166 params solve algorithmic tasks

dair_ai · x · 2026-11-17

Microsoft Research and collaborators introduce interpretable weight-space programming (iWSP): an LLM agent writes transformer weights and features explicitly in code with no gradient training, iterating 30–50 rounds of propose-test-revise, producing models readable line by line.

Paper: arXiv 2611.18114, code on GitHub.

Original post →

More from AGI Musings

AGI Musings channel →