open-source-vision-language-models-post-trained-rl-or-critique-tuning-act CLM-0463When open-source vision-language models are post-trained with RL or critique tuning to act as judges of chart-to-code outputs, they become more likely to output Accept rather than better at verification: pairwise-comparison accuracy barely improves under an order-swapped protocol, while acceptance rates rise and specificity falls, so they ratify flawed but plausible-looking charts.
Capability: Judging another's work as well as a careful reviewer would
Observed on
26 proprietary and open-source LMMs (7B-72B) judging chart reproduction and chart editing outputs on the 1,653-instance ChartJudgeBench; base-vs-post-trained comparisons within the.
Sources
- Measured: accuracy, F1, recall, specificity on 315 reproduction and 335 editing cases with near-balanced Accept/Reject labels from 5 expert annotators, plus base-vs-post-trained pairs (LLaVA-Critic-R1 98.1% recall / 9.4% specificity; edit accept rate 48.7% to 72.2%). Only two model families supply the before/after contrast, so the causal read on post-training is narrow; the perturbation-based CPA items are synthetic.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- When an 8B model verifies natural-language claims against knowledge-graph triples, scoring the evidence twice under opposing prompts (argue it supports, argue it refutes) and adjudicating between the two arguments avoids the bias toward a "Supported" verdict that a single-perspective assessment shows, and is the largest single contributor to accuracy in that pipeline.Checking claims against evidence · reviewed by AI
- General-purpose factuality verifiers (entailment-, consistency-, and RAG-trained) and zero-shot frontier LLM judges perform near chance at spotting unsupported claims in scientific peer reviews that must be checked against the full submitted paper, because they cannot separate paper-unsupported assertions from legitimate evaluative critique; small models fine-tuned on in-domain examples beat them by a wide margin.Stating false facts confidently · reviewed by AI
- On long-video (10-150 minute) audio-visual reasoning questions, multiple-choice accuracy overstates grounding: a large open model can answer about 55% of items correctly while scoring 8.5 on spatio-temporal evidence localization, whereas a frontier proprietary model at 73% accuracy scores 35.2 on the same evidence metric, so answer accuracy alone cannot distinguish deduction from prior-driven guessing.Reasoning about time in video · reviewed by AI
- For VLM-based judging of camera-conditioned world-model videos, splitting a long video pair into action-aligned chunks with structured visual evidence (source image, frame-grid overview, per-action first/last-frame panels) and voting chunk verdicts into video-level preferences agrees with human preferences better than judging the whole video pair at once, because short-lived action evidence is otherwise diluted by the full context.Judging another's work as well as a careful reviewer would · reviewed by AI
- General-purpose factuality verifiers (entailment-, consistency-, and RAG-trained) and zero-shot frontier LLM judges perform near chance at spotting unsupported claims in scientific peer reviews that must be checked against the full submitted paper, because they cannot separate paper-unsupported assertions from legitimate evaluative critique; small models fine-tuned on in-domain examples beat them by a wide margin.Stating false facts confidently · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A critique- or RL-post-trained open VLM judge that raises balanced accuracy and specificity on chart Accept/Reject verification over its base model, without an increased accept rate. Drafted stance toward used-steer-test-time-compute-generative-llm-judges-7b-70b-rerank: supports -- Both find that critique- or judge-oriented training of open models does not produce reliable evaluative signal.