REALISTA optimizes continuous combinations of valid editing directions in latent space to produce realistic adversarial prompts that elicit hallucinations more effectively than prior methods, including on large reasoning models.
Sophiavl-r1: Reinforcing mllms reasoning with thinking reward
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
representative citing papers
WebEye benchmark and Pixel-Searcher agent enable visual perception tasks by using web search to resolve object identities before precise localization or answering.
Video-R1 uses temporal-aware RL and mixed datasets to boost video reasoning in MLLMs, with a 7B model reaching 37.1% on VSI-Bench and surpassing GPT-4o.
PDCR improves vision-language reasoning by computing separate normalized confidence advantages for perception steps and reasoning steps after unsupervised decomposition.
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.
citing papers explorer
-
REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations
REALISTA optimizes continuous combinations of valid editing directions in latent space to produce realistic adversarial prompts that elicit hallucinations more effectively than prior methods, including on large reasoning models.
-
From Web to Pixels: Bringing Agentic Search into Visual Perception
WebEye benchmark and Pixel-Searcher agent enable visual perception tasks by using web search to resolve object identities before precise localization or answering.
-
PDCR: Perception-Decomposed Confidence Reward for Vision-Language Reasoning
PDCR improves vision-language reasoning by computing separate normalized confidence advantages for perception steps and reasoning steps after unsupervised decomposition.
-
Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.