Skip to main content
Search

# Model editing

3 related articles

Subscribe to updates
  1. arXiv

    A shared decision rule can now be corrected one owner at a time: 96.6% at 30 records, non-inferior to all seven baselines

    Several organisations apply the same published decision rule while each holds private records showing it is wrong for its own items, and none will pool them. One approach compiles the published rule exactly into a small ReLU network, lets each party submit only typed declarations, and gates them on evidence before compiling each into a weight patch confined to that party's reserved units. In two pre-registered synthetic tests, the learner scored 96.6% at 30 records per party against 91.7% for the unchanged rule and was non-inferior within 2 points to seven tuned baselines; the same search without the compiled prior scored 84.0%.

  2. arXiv

    Legal judgment models keep guilty verdicts after the decisive fact changes: PROSE cuts damage to supported rulings from 17.5% to 8.5%

    Legal judgment prediction models often keep a guilty verdict after a legally decisive condition has been altered, a tendency the authors call guilty bias. PROSE frames the correction as a paired constraint: the unsupported member of a case pair should flip, while its counterpart, differing in one decisive condition, should stay. On Qwen3-8B and GLM-4-9B it performs this selective repair through a one-time offline parameter edit, and at matched repair rates it reduces damage to supported verdicts from 17.5% to 8.5% on Qwen and from 24.3% to 7.1% on GLM.

  3. arXiv

    IWIBench lets large models rewrite a small model's weights: structured output scores zero across all ten editors, while alias edits hit the ceiling

    IWIBench turns gradient-free behavioural weight editing into a scored, controlled task: an editor model gets only four tools for reading and writing weights, running the target and inspecting it, and must make ten kinds of semantic change to Gemma 3 4B IT, after which fresh processes compare target success and control retention before and after the patch. Across ten configurations on the same thirty episodes, 6.1-sol at medium reasoning leads with a macro reward of 0.4601 and Astra follows at 0.4356; no editor scores on structured-response defaults, and only 5.6-terra earns positive reward on answer-length control.

You’re all caught up