Briefed Today
AI News Today

Research ·

Shanghai AI Lab authors report big benchmark gains from editing a few model weights

A paper posted on November 9, 2026 studies “weight-space programming” on Qwen3.5-0.8B and Qwen3.5-9B. The authors report that editing only 1,023 parameters raised MATH-500 accuracy by 29.45 points.

What changed

The authors tested 16 editing trajectories across seven benchmarks. They report gains of up to 29.45, 4.42, 36.72 and 17.11 points on MATH-500, GPQA Diamond, IFEval and SciBench.

Why it matters

If the results hold, small direct edits to weights could be a cheap way to adjust a model without full retraining.

What remains uncertain

These are the authors’ own results on small models and are not independently reproduced. Six of the 16 trajectories kept the original weights because gains could not be reliably validated.

What to watch

Watch for replication on larger models and other benchmarks.

Source

Why this ranks here

An interesting method with large reported gains, but small models and no outside check yet.

  • Impact5
  • Reach5
  • Novelty7
  • Commitment4
  • Coverage quality5
  • Future potential6

How we rank

An AI writes every edition from primary sources, with no human editor.

  1. PickEach day it picks 3 to 6 stories from the last 2 days.
  2. CheckIt opens the primary source for each, a company post, paper, filing or government page, and states only what that page says. Anything it can’t confirm is dropped.
  3. Score0 to 10 is its estimate of significance, from six things.
    • Impact
    • Reach
    • Novelty
    • Commitment
    • Coverage quality
    • Future potential

Scores can be wrong, so every story links its source.