AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Your Language Model Secretly Contains Personality Subnetworks

Ruimeng Ye, Zihan Wang, Zinan Ling, Yang Xiao, Manling Li, Xiaolong Ma, Bo Hui

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 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
add contrastive pruning to Wanda (ablation)Wanda persona pruning (Llama-2-13B) · AI Persona Wealth-Seekingaccuracy (%): 54.5 -> 66.0Improved
add contrastive pruning to Sparse (ablation)Sparse persona pruning (Llama-2-13B) · AI Persona Hallucinationaccuracy (%): 84.5 -> 96.0Improved
add contrastive pruning to Wanda (ablation)Wanda persona pruning (Llama-3-8B) · AI Persona Power-Seekingaccuracy (%): 57.0 -> 58.5No clear effect
use persona pruning instead of promptingLlama-2-13B · AI Persona Power-Seekingaccuracy (%): 41.0 -> 56.5Improved
use persona pruning instead of SFTLlama-2-13B · AI Persona Power-Seekingaccuracy (%): SFT 64.0 vs 56.5 best pruningHurt
use pruned subnetworks instead of PromptLlama-2-13B · RoleAgentBench Friendsaccuracy (%): 18.37 -> 41.84Improved
apply persona pruning to the base modelLlama-3-8B · MMLUaccuracy: 0.378 -> 0.362No clear effect
apply Sparse persona pruningLlama-3-8B · HellaSwagaccuracy: 0.675 -> 0.653No clear effect
use layer-aware MLP 0.3 sparsity instead of uniform 0.6 (ablation)Sparse pruning (Llama-2-13B) · MBTI persona-switchsuccess rate (%): 43.75 -> 56.25Improved
increase sparsity from 0.4 to 0.6 (ablation)Wanda pruning · MBTI persona switchsuccess rate (%): 68.75 -> 43.75Hurt
increase sparsity to 0.6 (ablation)Sparse pruning · MBTI persona switchsuccess rate (%): peak 75% at 0.6Improved
increase calibration samples from 5 to 100 (ablation)persona pruning · RoleAgentBench Sherlockaccuracy (%): 39.69 -> 50.79Improved

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