Study finds LLMs show modest loss aversion
The paper meta‑analyzes 102 estimates from 11 studies covering 32 language models to assess prospect theory components in large language models. The pooled loss‑aversion parameter λ is 1.26 (95 % CI 1.06 to 1.46), lower than the strongest human estimate of 1.96 and closer to conservative estimates of 1.07 and 1.31. Value‑function curvature is nearly linear (1.01) versus a human value of 0.67,…
1 source primary source
Key points
- Pooled λ = 1.26 (95% CI 1.06–1.46) for loss aversion
- Value function nearly linear at 1.01 versus human 0.67
- Probability weighting γ ≈0.81–0.84 inverse‑S shape
Read the original at arXiv cs.AI · by Adi Tarabeih primary sourceOpen source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- The Kaitchup releases statistically-lossless Qwen3.8-27B GGUF models · 1 src
- Moonshot AI announces Kimi K3.1 Pro with 1M-token context and 3D model creation · 2 src
- OpenAI’s GPT-6.1 Terra scores 1,671 points on Code Arena WebDev leaderboard · 1 src
- OpenAI and Anthropic cut AI model prices by up to 80% amid rising competition from open-weight models · 6 src
- OpenAI data shows 20.7% of ChatGPT messages seek information · 1 src
Comments
via GitHub Discussions