Skip to main content
Search

Jusheng Zhang

1 related articles

Subscribe to updates
  1. arXiv

    A language model edits 65 embedding rows in a robot policy, lifting success on reworded instructions from 74.3% to 87.5%

    GPT-6 Astra reads and rewrites the weights of a robot foundation policy directly, without gradients or new demonstrations, and locates one class of failure in the token-embedding layer. On the pi05_libero checkpoint, three independent editing sessions each improved held-out paraphrase success; the strongest changed only 65 embedding rows, raising success from 74.3% to 87.5% on the 335 instructions containing an edited token while the other 430 outcomes stayed unchanged. In 13 further sessions targeting other failures, the editor declined 12 candidate edits, so the demonstrated scope is lexical grounding repair.

You’re all caught up