Anchoring-LLMs
PublicCode, datasets, and replication notebook for the preprint Anchors in the Machine: Behavioral and Attributional Evidence of Anchoring Bias in LLMs. The project replicates and extends Tversky & Kahneman’s classic anchoring experiments across six open-source LLMs (GPT-2, GPT-Neo-125M, Falcon-RW-1B, Phi-2, Gemma-2B, LLaMA-2-7B).