Oriel — AI & technology worth noticing, every day.

The Social World Model Evaluation Gap

Paper · https://arxiv.org/abs/2610.33014

arXiv:2610.33014v1 Announce Type: new Abstract: A negotiation assistant can reach a good deal without accurately predicting the other person's response. We ask whether evaluations of social world models test that prediction directly. We audit 113 works published from 1999 to 2026. 9 primary evaluations measure a real person's response to an action selected by the system or experimenter. Only 2 also compare an explicit response forecast with that response. Their predictors use information available when the action is chosen, and the evaluated responses were excluded from fitting; both randomize the intervention. A separate full-text check of 25 language-model works finds no experiment meeting these conditions. The other evaluations test current-state inference, task success, or prediction from recorded actions. We propose a 1-step negotiation experiment that separates these questions. The findings describe the audited sample, not the prevalence of this evaluation gap across the field.

First seen by Oriel on .

AI summary

An audit of 113 social world model studies finds that direct evaluation of forecasts about real people's responses to chosen actions is rare, and proposes a negotiation experiment to test it.

  • Nine evaluations measure a real person's response to a system- or experimenter-selected action; only two also compare an explicit forecast with the observed response.
  • The two qualifying studies use information available when choosing the action, exclude evaluated responses from fitting, and randomize the intervention.
  • A full-text review of 25 language-model works finds no experiment meeting all these conditions.
  • Other evaluations measure current-state inference, task success, or predictions from recorded actions; the authors caution that their findings describe only the audited sample.

#social world models #evaluation #language models #negotiation

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
arXiv AI Papersoai:arXiv.org:2610.33014v1https://arxiv.org/abs/2610.33014 Read discussion