Research

Paper

AI LLM February 19, 2026

ADAPT: Hybrid Prompt Optimization for LLM Feature Visualization

Authors

João N. Cardoso, Arlindo L. Oliveira, Bruno Martins

Abstract

Understanding what features are encoded by learned directions in LLM activation space requires identifying inputs that strongly activate them. Feature visualization, which optimizes inputs to maximally activate a target direction, offers an alternative to costly dataset search approaches, but remains underexplored for LLMs due to the discrete nature of text. Furthermore, existing prompt optimization techniques are poorly suited to this domain, which is highly prone to local minima. To overcome these limitations, we introduce ADAPT, a hybrid method combining beam search initialization with adaptive gradient-guided mutation, designed around these failure modes. We evaluate on Sparse Autoencoder latents from Gemma 2 2B, proposing metrics grounded in dataset activation statistics to enable rigorous comparison, and show that ADAPT consistently outperforms prior methods across layers and latent types. Our results establish that feature visualization for LLMs is tractable, but requires design assumptions tailored to the domain.

Metadata

arXiv ID: 2602.17867
Provider: ARXIV
Primary Category: cs.LG
Published: 2026-02-19
Fetched: 2026-02-23 05:33

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.17867v1</id>\n    <title>ADAPT: Hybrid Prompt Optimization for LLM Feature Visualization</title>\n    <updated>2026-02-19T22:03:25Z</updated>\n    <link href='https://arxiv.org/abs/2602.17867v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.17867v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Understanding what features are encoded by learned directions in LLM activation space requires identifying inputs that strongly activate them. Feature visualization, which optimizes inputs to maximally activate a target direction, offers an alternative to costly dataset search approaches, but remains underexplored for LLMs due to the discrete nature of text. Furthermore, existing prompt optimization techniques are poorly suited to this domain, which is highly prone to local minima. To overcome these limitations, we introduce ADAPT, a hybrid method combining beam search initialization with adaptive gradient-guided mutation, designed around these failure modes. We evaluate on Sparse Autoencoder latents from Gemma 2 2B, proposing metrics grounded in dataset activation statistics to enable rigorous comparison, and show that ADAPT consistently outperforms prior methods across layers and latent types. Our results establish that feature visualization for LLMs is tractable, but requires design assumptions tailored to the domain.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-02-19T22:03:25Z</published>\n    <arxiv:primary_category term='cs.LG'/>\n    <author>\n      <name>João N. Cardoso</name>\n    </author>\n    <author>\n      <name>Arlindo L. Oliveira</name>\n    </author>\n    <author>\n      <name>Bruno Martins</name>\n    </author>\n  </entry>"
}