Paper

When Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New Tasks

Unreadunread

AI summary

arXiv:2606. 14629v1 Announce Type: new Abstract: Verifier-driven self-DPO is a common recipe for self-improving production visual-language models.

Reading status

Priority

About this paper

Authors: Jianzhe Lin
Released: 2026-06-15 (24d ago)
Topic: Related to current project
Relation to my work: Read next because When Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New Tasks overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, good, wrong, source, line, rate, recipe. Source: arxiv cs.CR (Cryptography and Security).
arXiv: 2606.14629

Ask Claude about this paper

Posts a comment with @claude so the runner spins up a Claude Code session scoped to When Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New Tasks. Replies appear in the conversation below.

Comments

No comments yet.

Related records

No links yet.