UPDATED TUE 03 NOV · 06:16 UTCLIVE
AILow

GPT-6 Astra edits model weights directly

Researchers say GPT-6 Astra changed 0.016% of Qwen3.5-0.8B's weights without training, improving math up to 6x.

What we know

An account on X RTs the claim, which the authors call Weight-Space Programming.

How the story developed

03 Nov · 06:16 UTC

Researchers say GPT-6 Astra changed 0.016% of Qwen3.5-0.8B's weights without training, improving math up to 6x.

02 Nov · 20:16 UTC

A claim that GPT-6 Astra edited 0.016% of Qwen3.5-0.8B's weights to improve math up to 6x without training drew reactions from prominent figures.

By the Global Eye 24 desk, with AI assistance. Updated as the story develops. Spot an error? Write to corrections@globaleye24.com.