Edit-Ready Training prepares vision models for LLM weight edits
Timo Ropinski and Sebastian Kranz presented “Training Vision Models for LLM-Driven Weight-Space Programming,” a benchmark covering five LLMs that edited convolutional networks and vision transformers across relabeling and error-correction tasks. The paper introduces Edit-Ready Training (ERT), which rehearses direct closed-form weight edits, and reports more than 2,600 edits scored for success, side effects and accuracy loss; ERT achieved the highest reported mean scores for GPT-6 Astra and Claude Opus 5.5 across the evaluated baselines.
The work directly evaluates whether dedicated training can make neural-network weights easier for LLMs to edit, while providing a concrete method and benchmark for studying this interaction.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2611.26408 (cs)
[Submitted on 19 Nov 2026]
Title:Training Vision Models for LLM-Driven Weight-Space Programming
Authors:Timo Ropinski, Sebastian Kranz
View a PDF of the paper titled Training Vision Models for LLM-Driven Weight-Space Programming, by Timo Ropinski and 1 other authors
#Edit-Ready Training#GPT-6 Astra#Claude Opus 5.5#Waterbirds#CIFAR-100#PACS#Entity-30