AI Paper Atlas

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

Easier to Judge than to Find: Predicting In-Context Learning Success for Demonstration Selection

Haochun Wang, Chaofen Yang, Jiatong Liu, Jingbo Wang, Zewen Qiang, Sendong Zhao, Bing Qin, Ting Liu

ICML 2026 · Large language models › Language, safety and interpretability · approach: In-context learning analysis · 0 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 DiSP judge-based selection instead of zero-shotLlama3-8B · 5-dataset averageaccuracy: 66.4 -> 77.6Improved
use DiSP judge-based selection instead of zero-shotQwen2.5-7B · 5-dataset averageaccuracy: 69.8 -> 82.9Improved
use DiSP instead of strongest baselineBM25 · Llama3-8B averageaccuracy: 74.8 -> 77.6Improved
use DiSP instead of strongest baselineBM25 · Qwen2.5-7B averageaccuracy: 80.5 -> 82.9Improved
use DiSP instead of Se2 on TRECSe2 · TREC, Llama3-8Baccuracy: 68.2 -> 79.2Improved
use learned selectors Se2 and SeDPO instead of zero-shotzero-shot · TREC, Llama3-8Baccuracy: 70.5 -> 67.5Hurt
add demonstrations to SST-2 querieszero-shot · SST-2, Llama3-8Baccuracy: 94.7 -> 95.4No clear effect
use DiSP over Uprise on training costUprise · training timeminutes: 13.4 -> 6.7Improved
use DiSP over SeDPO on training costSeDPO · training timeminutes: 157.1 -> 6.7Improved
train one pooled judge across all datasets instead of per-dataset judges (ablation)DiSP, l1 · Llama3-8Baccuracy: 91.5 -> 90.6Hurt
train one pooled judge across all datasets instead of per-dataset judges (ablation)DiSP, l2 · Llama3-8Baccuracy: 65.7 -> 60.5Hurt
train one pooled judge across all datasets instead of per-dataset judges (ablation)DiSP, l3 · Llama3-8Baccuracy: 55.8 -> 52.4Hurt

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