AI Paper Atlas

Research map / Large language models / Alignment and preferences

Contrastive Weak-to-Strong Generalization

Houcheng Jiang, Junfeng Fang, Jiaxin Wu, Tianyu Zhang, Chen Gao, Xiang Wang, Xiangnan He, Yang Deng

ICML 2026 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data · 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
add training-time contrast to ConG-S sampling to form ConG (ablation)ConG-S (w->s) · Qwen2.5-7B weak-to-strongAvg. of AlpacaEval2 and Arena-Hard win rates: 48.0 -> 52.7Improved
add training-time contrast to ConG-S sampling to form ConG (ablation)ConG-S (w->s) · Llama3-8B weak-to-strongAvg. of AlpacaEval2 and Arena-Hard win rates: 36.7 -> 41.6Improved
use ConG instead of SimPO in self-alignmentSimPO · Qwen2.5-3B self-alignmentAvg. of AlpacaEval2 and Arena-Hard win rates: 41.5 -> 45.5Improved
use ConG instead of SimPO in self-alignmentSimPO · Llama3.2-3B self-alignmentAvg. of AlpacaEval2 and Arena-Hard win rates: 33.5 -> 34.6Improved
use ConG instead of WSPO for weak-to-strongWSPO · Qwen2.5-7B weak-to-strongAvg. of AlpacaEval2 and Arena-Hard win rates: 25.6 -> 52.7Improved
use ConG instead of Weak SFT (post)Weak SFT (post) · Llama3-8B weak-to-strongAvg. of AlpacaEval2 and Arena-Hard win rates: 27.2 -> 41.6Improved
use ConG instead of WeakTeacher (weak model outputs)WeakTeacher · Qwen2.5-7B weak-to-strongAvg. of AlpacaEval2 and Arena-Hard win rates: 43.5 -> 52.7Improved
use cross-family weak model (Llama3.2-3B) instead of same-familyConG · Qwen2.5-7B strong modelAvg. of AlpacaEval2 and Arena-Hard win rates: 52.7 -> 51.7No clear effect
use GPT-5 judge instead of GPT-4ConG · Qwen2.5-7B weak-to-strongAvg. of AlpacaEval2 and Arena-Hard win rates: 52.7 -> 47.3No clear effect
apply ConG instead of Base modelQwen2.5-7B-Instruct · GSM8Kaccuracy: 80.44 -> 80.21No clear effect
apply ConG instead of Base modelLlama3-8B-Instruct · GSM8Kaccuracy: 75.28 -> 75.66No clear effect
apply Weak SFT instead of Base modelQwen2.5-7B-Instruct · GSM8Kaccuracy: 80.44 -> 75.28Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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