Entity · paper

Gotta Catch them all: the modes of Sycophancy

paperactiveprovisionalgotta-catch-them-all-the-modes-of-sycophancy-8318341e·1 events·first seen 4d ago

Aliases: Gotta Catch them all: the modes of Sycophancy

More like this (12)

SycophancyEval sycophancy Sycophantic Praise: Evaluating Excessive Praise in Language Models Supra Cognitive Modes Recalling Too Well: Sycophancy Evaluation and Mitigation in Memory-Augmented Models CATCH-ME Beyond Sycophancy: Structured Resistance and Compliance in LLM Moral Reasoning How Does Alignment Tuning Shape Representations of Sycophancy and Related Cue-Induced Biases in LLMs?Competitive Self-Play Decomposing Factual Sycophancy in Language Models: How Size and Instruction Tuning Shape Robustness Skill Self-Play Faithfulness to Refusal: A Causal Audit of Neuron Selectors

Recent events (1)

6arXiv · cs.CL·4d ago·source ↗

Mechanistic analysis reveals sycophancy in LLMs is a structured family of distinct modes, not a single behavior

A new arXiv preprint challenges the assumption that sycophancy in LLMs is a monolithic tendency by analyzing three hypothesized modes across 948 social pressure situations. Despite producing nearly identical outputs (text classifier accuracy 57.8%), the modes are perfectly linearly separable in internal representations from layer 14 onward, emerge at different processing stages, and rely on distinct attention circuitry. The findings suggest that sycophancy measurement and intervention strategies need to be more precise and mode-specific.

Evaluation and Benchmarking AI Safety Research Gotta Catch them all: the modes of Sycophancy +1 more