AI Paper Atlas

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

High Accuracy, Less Talk (HALT): Reliable LLMs through Capability-Aligned Finetuning

Tim Franzmeyer, Archie Sravankumar, Lijuan Liu, Yuning Mao, Rui Hou, Sinong Wang, Jakob Nicolaus Foerster, Luke Zettlemoyer, Madian Khabsa

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 1 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
finetune on HALT capability-aligned data instead of unchanged data (n_capable)Llama3-70B · WikibiosF1: 0.35 -> 0.68Improved
finetune on HALT data instead of unchanged data (n_capable)Llama3-70B · MATHF1: 0.75 -> 0.87Improved
finetune on HALT data instead of unchanged data (n_capable)Llama3-8B · WikibiosF1: 0.26 -> 0.63Improved
finetune on HALT data vs unchanged (n_all)Llama3-70B · APPS introF1: 0.44 -> 0.41Hurt
finetune on HALT data vs unchanged (n_all)Mistral-7B · WikibiosF1: 0.14 -> 0.13Hurt
replace removal by capability-aligned removal vs RandomTrim of same length (ablation)Llama3-70B · WikibiosF1 (n_all): 0.12 -> 0.33Improved
use HALT instead of IDK abstention (ablation)Llama3-70B · APPS introF1 (n_all): 0.19 -> 0.41Improved
train combined multi-domain model with HALT alpha=40%Llama3-70B · four domainscorrectness (%): 51 -> 87Improved
train combined multi-domain model with HALT alpha=40%Llama3-70B · four domainscompleteness (%): 51 -> 25Hurt
lower alpha percentile of selected preliminary response (ablation)HALT Llama3-70B · four datasetscorrectness: +-17%Improved
HALT-finetune vs base modelLlama3-70B · AlpacaEval 578 promptswin rate: 50.8%No clear effect
HALT-finetune vs base modelLlama3-70B · Wikibios AlpacaEvalwin rate: 66.0%Improved

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