AI Paper Atlas

Research map / Large language models / Alignment and preferences

Controllable and Explainable Personality Sliders for LLMs at Inference Time

Florian Hoppe, David Khachaturov, Robert D. Mullins, Mark Huasong Meng

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
train subsequent probes on activations shifted by prior probes (Sequential Adaptive Steering) instead of independent probes (ablation)multi-trait activation steering · E+ N+ A- profileperplexity: 27.16 -> 14.54Improved
train subsequent probes on shifted activations (ablation)multi-trait steering · E+ N+ A- profileExtraversion score: 3.62 -> 4.20Improved
train subsequent probes on shifted activations (ablation)multi-trait steering · E+ N+ A- profileNeuroticism score: 3.10 -> 4.1Improved
add Fisher-ratio automated layer selection instead of random layers (ablation)SAS · E+ N+ A- profileExtraversion score: 3.30 -> 4.20Improved
add Fisher-ratio layer selection instead of random layers (ablation)SAS · E+ N+ A- profileperplexity: 14.79 -> 14.54No clear effect
replace independent probes with Gram-Schmidt orthogonalization post hocnaive steering · perplexity at alpha 1.0perplexity: 17.10 -> 17.15No clear effect
use SAS instead of naive or Gram-Schmidt at alpha 1.5multi-trait steering · perplexity at alpha 1.5perplexity: > 83 -> 18.65Improved
use SAS instead of LoRA Soups merging three adaptersmulti-trait personality control · E+ N+ A- profiletrait shift: Extraversion -0.12 -> +1.00Improved
use SAS instead of DPO fine-tuned modelmulti-trait personality control · E+ N+ A- profiletrait shiftImproved
add SAS steeringLlama-3-8B-Instruct · TruthfulQA and HellaSwagaccuracy: 0.32 -> 0.32; 0.64 -> 0.64No clear effect
add SAS steeringLlama-3-8B-Instruct · generation qualityperplexity: 9.95 -> 14.54Hurt
orthogonalize probes with SAS trainingpersonality steering vectors · inter-probe cosine similarityoff-diagonal cosine similarityImproved

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