chameleon-dual-behavior-chameleon-sloth

IN premise — summaries/2026/08/24/xie-2024-chameleon-sloth-s0-abstract.md

Created 2026-08-25T02:58:59+00:00

LLMs exhibit dual behavior: high receptiveness to coherent single-source counter-memory (chameleon-like) but strong confirmation bias preferring parametric memory when both supporting and conflicting evidence are present simultaneously (sloth-like).

Summary

LLMs are easy to sway when you present a single, well-organized piece of contradicting information in isolation, but when both sides of a conflict are on the table at the same time, they fall back on whatever they already "know" from training. This asymmetry matters because it means the model can be reliably updated by focused counter-evidence, yet will resist genuine revision when a fuller picture is shown, which complicates any system that needs it to accurately weigh competing sources.