Research map / Large language models / Language, safety and interpretability
Capturing Failures of Large Language Models via Human Cognitive Biases
Erik Jones, Jacob Steinhardt
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 |
|---|---|---|---|
| prepend irrelevant preceding function (framing line) to prompt | Codex · HumanEval | functional accuracy: -22.3 to -30.5 points | Hurt |
| prepend irrelevant preceding function | Codex · HumanEval | framing line output rate: 4.5% -> 81% | Hurt |
| prepend irrelevant preceding function | CodeGen · HumanEval | framing line output rate: 0.0% -> 70.7% | Hurt |
| flip operation order from unary-first to binary-first | Codex · MathEquations | accuracy: 50% -> 17% | Hurt |
| request a contradictory function name | Codex · MathEquations | accuracy: 100% -> 4.4% | Hurt |
| add anchoring lines from canonical solution to prompt | Codex · HumanEval | anchor-based generation: for var in 32%-61% of solutions | Hurt |
| anchor numerical estimate prompts on GPT-3 | GPT-3 · anchoring adaptation | estimate shift: no change 10.7 vs 14.3 for 20% vs 50% anchors | Mixed |
Closest papers by meaning
- Human Feedback is not Gold Standard (ICLR 2024 · similarity 0.88)
- Gaining Wisdom from Setbacks: Aligning Large Language Models via Mistake Analysis (ICLR 2024 · similarity 0.87)
- Training on the Test Task Confounds Evaluation and Emergence (ICLR 2025 · similarity 0.87)
- Attention Satisfies: A Constraint-Satisfaction Lens on Factual Errors of Language Models (ICLR 2024 · similarity 0.87)
- CogBench: a large language model walks into a psychology lab (ICML 2024 · similarity 0.87)
- Do Large Language Models Perform the Way People Expect? Measuring the Human Generalization Function (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.
