Paper Meta-Analyzes Prospect Theory Features of Large Language Models
Get up to speed
Researcher Adi Tarabeih released a meta-analysis paper summarizing 102 fitting estimates from 11 studies covering 32 large language models, examining whether large language models exhibit prospect theory characteristics consistent with humans. Results show that large language models have a lower degree of loss aversion, with a pooled λ value of 1.26, far below the strongest human meta-analysis estimate of 1.96. The diminishing sensitivity of their value function varies by measurement method, while probability weighting exhibits an inverse S-shaped pattern similar to humans.
AI-generated from reports · 2 hr agoUpdated
Coverage timeline
Follow the coverage to understand different sides of the event.
- arXiv q-finProspect Theory in Large Language Models: A Meta-Analysis of Loss Aversion, Diminishing Sensitivity, and Probability Weighting
A meta-analysis of 102 fitted estimates across 11 studies covering 32 LLMs shows low loss aversion, with a pooled λ of 1.26, far below the strongest human meta-analytic estimate of 1.96. Diminishing sensitivity of the value function varies by measurement method, but probability weighting exhibits a human-like inverse-S shape.
Trend of this event
Not enough consecutive observations yet; trend is not drawn.