Research map / Large language models / Alignment and preferences
Expectation Alignment of Language Models for Real-World User Expectations
Miaomiao Li, Yang Wang, Bin Liang, Shudong Liu, Zhiwei Zhang, Kam-Fai Wong
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 LENS latent expectation projector | LLaMA-3.1-8B · ExpectBench overall | expectation satisfaction: 2.17 -> 2.37 | Improved |
| add LENS latent expectation projector | Mistral-7B · ExpectBench overall | expectation satisfaction: 2.39 -> 2.50 | Improved |
| supervised fine-tuning on gold expectations (SFT-Exp) instead of frozen model (ablation) | LLaMA-3.1-8B · ExpectBench overall | expectation satisfaction: 2.17 -> 2.21 | No clear effect |
| supervised fine-tuning on gold expectations (SFT-Exp) (ablation) | Mistral-7B · ExpectBench overall | expectation satisfaction: 2.39 -> 2.39 | No clear effect |
| fine-tune on responses (SFT-Resp) (ablation) | LLaMA-3.1-8B · ExpectBench overall | expectation satisfaction: 2.17 -> 2.33 | Improved |
| fine-tune on responses (SFT-Resp) (ablation) | Mistral-7B · ExpectBench overall | expectation satisfaction: 2.39 -> 2.44 | Improved |
| use LENS instead of SFT-Resp (ablation) | LLaMA-3.1-8B · ExpectBench overall | expectation satisfaction: 2.33 -> 2.37 | Improved |
| use LENS instead of SFT-Resp (ablation) | Mistral-7B · ExpectBench overall | expectation satisfaction: 2.44 -> 2.50 | Improved |
| use LENS instead of SFT-Resp on Cx dimension (ablation) | LLaMA-3.1-8B · ExpectBench dimension Cx | expectation satisfaction: 2.36 -> 2.41 | Improved |
| use LENS instead of SFT-Resp on Cr dimension (ablation) | LLaMA-3.1-8B · ExpectBench dimension Cr | expectation satisfaction: 2.29 -> 2.29 | No clear effect |
| increase prediction budget k up to 10 | six LLMs · Expectation prediction coverage | coverage | Improved |
Closest papers by meaning
- Do Large Language Models Perform the Way People Expect? Measuring the Human Generalization Function (ICML 2024 · similarity 0.89)
- Peering Through Preferences: Unraveling Feedback Acquisition for Aligning Large Language Models (ICLR 2024 · similarity 0.88)
- Understanding the Learning Dynamics of Alignment with Human Feedback (ICML 2024 · similarity 0.88)
- Accelerating Unbiased LLM Evaluation via Synthetic Feedback (ICML 2025 · similarity 0.87)
- Aligning to Thousands of Preferences via System Message Generalization (NeurIPS 2024 · similarity 0.87)
- Fundamental Limitations of Alignment in Large Language Models (ICML 2024 · 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.
