Research map / Large language models / AI and society
Using cognitive models to reveal value trade-offs in language models
Sonia Krishna Murthy, Rosie Zhao, Jennifer Hu, Sham M. Kakade, Markus Wulfmeier, Peng Qian, Tomer Ullman
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add low or medium reasoning effort | closed-source LLMs · Polite speech task | projected informational mixture phi: beta_low = 0.228, beta_medium = 0.211 | Improved |
| increase reasoning effort from low to medium | closed-source LLMs · Polite speech task | phi | No clear effect |
| prompt for social goal | closed-source LLMs · Polite speech task | w_pre, w_inf, phi | Mixed |
| prompt for informative goal | closed-source LLMs · Polite speech task | w_inf and phi | Improved |
| use Claude-Sonnet-3.7 low reasoning instead of Claude-Sonnet-3.5 no reasoning | Claude · Polite speech task | w_inf: t = -5.49, p = 0.005 | Improved |
| add reasoning to Gemini-Flash | Gemini-Flash · Polite speech task | w_inf, w_soc, w_pre: p = 0.43 for w_inf, p = 0.20 for w_soc, p = 0.89 for w_pre | No clear effect |
| use UltraFeedback instead of HH-RLHF as feedback dataset (ablation) | Qwen and Llama aligned with DPO/PPO · Polite speech task | w_inf | Improved |
| use HH-RLHF instead of UltraFeedback (ablation) | Qwen and Llama aligned with DPO/PPO · Polite speech task | w_soc | Improved |
| use PPO instead of DPO (ablation) | Qwen and Llama · Polite speech task | phi: approx. 0.7 | Mixed |
| use Qwen-Instruct instead of Llama-Instruct base (ablation) | open-source alignment · Polite speech task | w_inf, w_pre | Mixed |
Closest papers by meaning
- How do Large Language Models Navigate Conflicts between Honesty and Helpfulness? (ICML 2024 · similarity 0.89)
- Language Models Trained to do Arithmetic Predict Human Risky and Intertemporal Choice (ICLR 2025 · similarity 0.88)
- Do LLMs have Consistent Values? (ICLR 2025 · similarity 0.88)
- Large Language Models Assume People are More Rational than We Really are (ICLR 2025 · similarity 0.87)
- Distributive Fairness in Large Language Models: Evaluating Alignment with Human Values (NeurIPS 2025 · similarity 0.87)
- Value Entanglement: Conflation Between Different Kinds of Good In (Some) Large Language Models (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
