chameleon-order-sensitivity-model-dependent

IN premise — summaries/2026/08/24/xie-2024-chameleon-sloth-s0-abstract-chunk-2.md

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

Evidence order sensitivity is model-dependent: ChatGPT favors first-presented evidence, while PaLM2 and Llama2-7B favor later-presented evidence with fluctuations exceeding 30%.

Summary

Different AI models get pulled in different directions depending on whether evidence is presented first or last, and the swings can be over 30 percent. This means the order in which you feed information to a model is not a neutral choice — it can materially change what conclusion it reaches, and you need to account for that bias per model when structuring prompts or evidence sequences.