AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Ice Cream Doesn’t Cause Drowning: Benchmarking LLMs Against Statistical Pitfalls in Causal Inference

Jin Du, Li Chen, Xun Xian, An Luo, Fangqiao Tian, Ganghua Wang, Charles Doss, Xiaotong Shen, Jie Ding

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 7 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use code-assisted prompting instead of direct promptingGPT-4.1 · CausalPitfallscausal reliability: 25.24 -> 37.32Improved
Use code-assisted prompting instead of direct promptingGemini-2.0-flash · CausalPitfallscausal reliability: 24.36 -> 31.80Improved
Use code-assisted prompting instead of direct promptingGPT-o4-mini · CausalPitfallscausal reliability: 40.72 -> 43.03Improved
Use code-assisted prompting instead of direct promptingMistral-7b · CausalPitfallscausal reliability: 14.43 -> 7.65Hurt
Use code-assisted prompting instead of direct promptingLlama3.1-8b · CausalPitfallscausal reliability: 17.86 -> 12.86Hurt
Use code-assisted prompting instead of direct promptingGemma2-9b · CausalPitfallscausal reliability: 13.69 -> 13.24No clear effect
Use code-assisted prompting instead of direct promptingGPT-o4-mini · CausalPitfalls very hard questionscausal reliability: 17.75 -> 32.84Improved
Use code-assisted prompting instead of direct promptingGPT-o4-mini · CausalPitfalls very easy questionscausal reliability: 60.72 -> 56.73Hurt
Remove hints from question (very easy to very hard)GPT-o4-mini direct prompting · CausalPitfallscausal reliability: 60.72 -> 17.75Hurt
Change brand label from HealthPlus to UltraSugar with identical dataGPT-4o and Gemini-2.0-flash · Branding bias testconclusion correctnessHurt
Allow one debugging attempt for code-assisted promptingMistral-7B and Llama-8B · CausalPitfallscausal reliabilityImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy