# Toward Latent Language Model Skills Steering and Optimization: An Empirical Study

arXiv: https://arxiv.org/abs/2608.29459 · HTML: https://mlsanity.com/paper/2608.29459/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 32 references that are in the map.
- First posted on arXiv 29 Aug 2026.

## Builds on (most cited first, 23 of 23)

- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [Persona Vectors: Monitoring and Controlling Character Traits in Language Models](https://mlsanity.com/paper/2507.21509/) (2025)
- [SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning](https://mlsanity.com/paper/2602.08234/) (2026)
- [Programming Refusal with Conditional Activation Steering](https://mlsanity.com/paper/2409.05907/) (2024)
- [SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills](https://mlsanity.com/paper/2504.07079/) (2025)
- [MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents](https://mlsanity.com/paper/2602.02474/) (2026)
- [Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimization](https://mlsanity.com/paper/2406.00045/) (2024)
- [Learning Graph Embeddings for Compositional Zero-shot Learning](https://mlsanity.com/paper/2102.01987/) (2021)
- [EvoSkill: Automated Skill Discovery for Multi-Agent Systems](https://mlsanity.com/paper/2603.02766/) (2026)
- [A causal view of compositional zero-shot recognition](https://mlsanity.com/paper/2006.14610/) (2020)
- [Instruct-SkillMix: A Powerful Pipeline for LLM Instruction Tuning](https://mlsanity.com/paper/2408.14774/) (2024)
- [Improved Representation Steering for Language Models](https://mlsanity.com/paper/2505.20809/) (2025)
- [SAKE: Steering Activations for Knowledge Editing](https://mlsanity.com/paper/2503.01751/) (2025)
- [EasySteer: A Unified Framework for High-Performance and Extensible LLM Steering](https://mlsanity.com/paper/2509.25175/) (2025)
- [Automated Skill Discovery for Language Agents through Exploration and Iterative Feedback](https://mlsanity.com/paper/2506.04287/) (2025)
- [CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features](https://mlsanity.com/paper/2508.12535/) (2025)
- [PERSONA: Dynamic and Compositional Inference-Time Personality Control via Activation Vector Algebra](https://mlsanity.com/paper/2602.15669/) (2026)
- [BILLY: Steering Large Language Models via Merging Persona Vectors for Creative Generation](https://mlsanity.com/paper/2510.10157/) (2025)
- [How Does RL Post-training Induce Skill Composition? A Case Study on Countdown](https://mlsanity.com/paper/2512.01775/) (2025)
- [Towards Compositional Generalization of LLMs via Skill Taxonomy Guided Data Synthesis](https://mlsanity.com/paper/2601.03676/) (2026)
- [Scaling laws for activation steering with Llama 2 models and refusal mechanisms](https://mlsanity.com/paper/2507.11771/) (2025)

## Abstract (arXiv)

Skills, as a useful abstraction for the procedural capabilities of large language models (LLMs), capture how models perform structured, multi-step reasoning and program execution. Existing approaches typically treat skills as explicit, surface-level constructs specified through prompts or programs, leaving open the question of how such procedural capabilities are represented inside the model and whether they can be manipulated as structured objects in latent space. In this empirical study, we investigate whether procedural LLM skills can be represented as directions in activation space and whether vector-space operations over these directions can express skill-level behaviors. We find that procedural skills admit a vector-space representation: individual skill directions can be activated to shift model behavior; independently extracted directions can compose to form higher-level skills. Contrastive directions yield context-conditioned algorithmic personalization and optimization trajectories over skill directions evolve non-monotonically, with intermediate states often surpassing fully optimized solutions. These results support a representation-level view of procedural LLM skills: they admit a latent vector-space organization that allows direct manipulation through internal interventions.