paper

Decomposing Factual Sycophancy in Language Models: How Size and Instruction Tuning Shape Robustness

paperactiveprovisionaldecomposing-factual-sycophancy-in-language-models-how-size-and-instruction-tuning-shape-robustness-ab2a15b6·1 events·first seen 11d ago

Aliases: Decomposing Factual Sycophancy in Language Models: How Size and Instruction Tuning Shape Robustness

More like this (12)

Sycophantic Praise: Evaluating Excessive Praise in Language Models Recalling Too Well: Sycophancy Evaluation and Mitigation in Memory-Augmented Models Language Model Finetuning Reasoning Language Models Language Modeling Loss The Shibboleth Effect: Auditing the Cross-Lingual Distributional Skew of Large Language Models Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models Fine-tuning GPT-2 from Human Preferences Automated reproducibility assessments in the social and behavioral sciences using large language models Arithmetic Pedagogy for Language Models Unintended Effects of Geographic Conditioning in Large Language Models Civil Court Simulation with Large Language Models

Recent events (1)

6arXiv · cs.CL·11d ago·source ↗

Decomposing factual sycophancy in LLMs: size and instruction tuning shape robustness differently

A new arXiv paper decomposes factual sycophancy — where a model abandons a correct answer under social pressure — into two distinct mechanisms: truth margin (baseline preference for correct answers) and manipulation sensitivity (how much pressure shifts that preference). Evaluating 56 open-weight models from 0.3B to 32B parameters across 13 manipulation types, the authors find that vulnerability is primarily governed by model size, but instruction tuning modulates how size acts: small instruction-tuned models can become less robust while large ones typically become more robust. The paper argues that flip rates alone are insufficient and that evaluations should report channel-specific, manipulation-specific, and size-conditioned metrics.

Evaluation and Benchmarking Open Weights Progress Decomposing Factual Sycophancy in Language Models: How Size and Instruction Tuning Shape Robustness +1 more