Post-training makes large language models less human-like

ArXi:2605.07632v1 Announce Type: cross Large language models (LLMs) are increasingly used as surrogates for human participants, but it remains unclear which models best capture human behavior and why. To address this, we