AI Paper Atlas

Research map / Large language models / Alignment and preferences

ALPHA: Action-Based Learning for Pluralistic Human Alignment in Large Language Models

Aanisha Bhattacharyya, Susmit Agrawal, Yaman Kumar Singla, Tarun Ram Menta, Nikitha Sr, Rajiv Ratn Shah, Changyou Chen, Balaji Krishnamurthy

AAAI 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior · 3 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
fine-tune chat model on ALPHA50M behavioral dataLlama-2-7B-chat · OpinionQA-XLrepresentativeness: 83.61 -> 85.15Improved
fine-tune chat model on ALPHA50M behavioral dataLlama-2-70B-chat · GSSsteerability: 77.18 -> 81.30Improved
fine-tune chat model on ALPHA50M behavioral dataLlama-2-13B-chat · OpinionQA-XLsteerability: 81.11 -> 83.54Improved
fine-tune on ALPHA50M instead of task-specific CultureBank SFTLlama-2-7B · CultureBank Redditentailment score: 85.93 -> 92.39Improved
fine-tune chat model on ALPHA50M behavioral dataLlama-3-8B-chat · CultureBank Redditentailment score: 85.38 -> 85.38No clear effect
fine-tune on ALPHA50MLlama-2-13B-chat · post-cutoff Pew surveys Mar-Aug 2023opinion alignment: 82.97 -> 84.72Improved
fine-tune on ALPHA50MLlama-3-8B-Instruct · post-cutoff Pew surveys Mar-Aug 2023opinion alignment: 83.65 -> 84.89Improved
train on single behavior task (Transsuasion or TAG) instead of all tasks (ablation)ALPHA-13B · OpinionQA-XLrepresentativeness: 84.82 and 84.57 vs 86.65 fullHurt
add TAP task for cultural generalization (ablation)ALPHA-13B · CultureNLI INentailment score: 68.39 (best single task)Improved
vary which behavior task is used (ablation)ALPHA-13B · GSSsteerabilityNo clear effect

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