Research map / Large language models / Alignment and preferences
The Wisdom of Hindsight Makes Language Models Better Instruction Followers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use hindsight instruction relabeling rather than FARL's correct-answer-only filtering | FLAN-T5-large · Tracking Shuffled Objects (5) | accuracy: 15.6 -> 61.2 | Helped |
| Use hindsight instruction relabeling rather than FARL's correct-answer-only filtering | FLAN-T5-large · Logical Deduction (5 Objects) | accuracy: 54.0 -> 67.0 | Helped |
| Use hindsight instruction relabeling rather than FARL's correct-answer-only filtering | FLAN-T5-large · Word Sorting | accuracy: 3.4 -> 3.4 | No clear effect |
| Add sub-output sampling (ablation) | HIR · Geometric Shapes; Tracking Shuffled Objects (3); Logical Deduction (3 Objects) | accuracy: 86.1 -> 90.3; 100.0 -> 100.0; 75.0 -> 91.7 | Helped |
| Add entropy regularization (ablation) | HIR · Geometric Shapes; Tracking Shuffled Objects (3); Logical Deduction (3 Objects) | accuracy: 47.2 -> 90.3; 100.0 -> 100.0; 48.3 -> 91.7 | Helped |
| Add label smoothing (ablation) | HIR · Geometric Shapes; Tracking Shuffled Objects (3); Logical Deduction (3 Objects) | accuracy: 84.7 -> 90.3; 100.0 -> 100.0; 23.3 -> 91.7 | Helped |
Closest papers by meaning
- Checklists Are Better Than Reward Models For Aligning Language Models (NeurIPS 2025 · similarity 0.90)
- Chain of Hindsight aligns Language Models with Feedback (ICLR 2024 · similarity 0.90)
- ReMax: A Simple, Effective, and Efficient Reinforcement Learning Method for Aligning Large Language Models (ICML 2024 · similarity 0.89)
- BLEUBERI: BLEU is a surprisingly effective reward for instruction following (NeurIPS 2025 · similarity 0.89)
- Incentivizing Reasoning for Advanced Instruction-Following of Large Language Models (NeurIPS 2025 · similarity 0.89)
- Generalizing Verifiable Instruction Following (NeurIPS 2025 · similarity 0.89)
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.