VISE is the first benchmark for sycophancy in Video-LLMs, with two training-free mitigation strategies based on key-frame selection and internal representation steering.
findings-emnlp.370/
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
Self-ReSET is a reinforcement learning approach that lets large reasoning models learn to recover from their own unsafe reasoning trajectories, improving robustness to adversarial jailbreaks while preserving utility.
By injecting arithmetic mistakes into CoT reasoning, the paper identifies a hidden critique ability in LRMs and extracts a steerable critique vector that enhances self-correction across model scales.
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
citing papers explorer
-
Flattery in Motion: Benchmarking and Analyzing Sycophancy in Video-LLMs
VISE is the first benchmark for sycophancy in Video-LLMs, with two training-free mitigation strategies based on key-frame selection and internal representation steering.
-
Self-ReSET: Learning to Self-Recover from Unsafe Reasoning Trajectories
Self-ReSET is a reinforcement learning approach that lets large reasoning models learn to recover from their own unsafe reasoning trajectories, improving robustness to adversarial jailbreaks while preserving utility.
-
Decoding the Critique Mechanism in Large Reasoning Models
By injecting arithmetic mistakes into CoT reasoning, the paper identifies a hidden critique ability in LRMs and extracts a steerable critique vector that enhances self-correction across model scales.
-
RetroInfer: A Vector Storage Engine for Scalable Long-Context LLM Inference
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.