Hierarchical Experimentalist Agents

Hugging Face Daily Papers Papers
active-experimentation self-improvement llm-agents physics-benchmark skill-learning

Summary

Introduces HExA, a training-free framework enabling LLMs to learn through active experimentation and skill reuse, achieving up to 77% success on the new Interphyre physics benchmark, a large improvement over existing agents.

Large language models (LLMs) are increasingly used to take actions in the real world and support human decision-making, yet most agents rely on parametric knowledge, fixed post-training data, retrieval, or search. This paradigm breaks down in novel domains and for sophisticated queries that cannot be answered from prior knowledge alone. Knowing the laws of physics, for instance, does not by itself enable LLMs to answer queries or complete long-horizon tasks in a complex physical system. To address this, we introduce Hierarchical Experimentalist Agents (HExA), an in-context self-improvement framework to learn from active experimentation. HExA iteratively designs and refines query-relevant experiments, learns a reusable library of composable skills from experience, and integrates experimental evidence to answer queries or take actions. HExA is training-free, compatible with any black-box model, and does not require external supervision, oracles, or offline data. To evaluate active experimentation, we introduce Interphyre, a tool-calling benchmark built on the PHYRE 2D procedural physics environment, where agents propose interventions and test hypotheses through simulation APIs. Experiments show that current LLM agents struggle in these settings, especially on the hardest levels of Interphyre. Claude Sonnet 4.6 achieves only 2% success, while HExA improves the same model to up to 77% success. HExA also improves open-weight models and outperforms agentic baselines such as ReAct and Reflexion. Moreover, using only skills learned from easier levels and transferred without active experimentation, HExA achieves 44% success, demonstrating the reusability and generalization of its learned skills. Overall, HExA shows that learning through active experimentation can help agents discover useful knowledge, acquire reusable skills, and make efficient progress on novel long-horizon tasks.
Original Article
View Cached Full Text

Cached at: 07/01/26, 07:44 PM

Paper page - Hierarchical Experimentalist Agents

Source: https://huggingface.co/papers/2606.29315

Abstract

HExA enables large language models to improve through active experimentation and skill learning in novel domains without requiring training or external supervision.

Large language models (LLMs) are increasingly used to take actions in the real world and support human decision-making, yet most agents rely on parametric knowledge, fixed post-training data, retrieval, or search. This paradigm breaks down in novel domains and for sophisticated queries that cannot be answered from prior knowledge alone. Knowing the laws of physics, for instance, does not by itself enable LLMs to answer queries or complete long-horizon tasks in a complex physical system. To address this, we introduceHierarchical Experimentalist Agents(HExA), an in-contextself-improvement frameworkto learn fromactive experimentation. HExA iteratively designs and refinesquery-relevant experiments, learns areusable library of composable skillsfrom experience, and integratesexperimental evidenceto answer queries or take actions. HExA is training-free, compatible with any black-box model, and does not require external supervision, oracles, or offline data. To evaluateactive experimentation, we introduceInterphyre, atool-calling benchmarkbuilt on thePHYRE 2D procedural physics environment, where agents propose interventions and test hypotheses throughsimulation APIs. Experiments show that current LLM agents struggle in these settings, especially on the hardest levels ofInterphyre. Claude Sonnet 4.6 achieves only 2% success, while HExA improves the same model to up to 77% success. HExA also improves open-weight models and outperforms agentic baselines such asReActandReflexion. Moreover, using only skills learned from easier levels and transferred withoutactive experimentation, HExA achieves 44% success, demonstrating the reusability and generalization of its learned skills. Overall, HExA shows that learning throughactive experimentationcan help agents discover useful knowledge, acquire reusable skills, and make efficient progress on novel long-horizon tasks.

View arXiv pageView PDFProject pageGitHub0Add to collection

Get this paper in your agent:

hf papers read 2606\.29315

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.29315 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.29315 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.29315 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Hierarchical Experimentalist Agents

arXiv cs.AI

Introduces Hierarchical Experimentalist Agents (HExA), an in-context, experiment-centric self-improvement framework that enables LLM agents to design experiments, learn reusable skills, and answer queries in novel domains, achieving significant improvements over baselines on the Interphyre physics simulation benchmark.

Toward Auditable AI Scientists: A Hypothesis Evolution Protocol for LLM Agents

arXiv cs.AI

This paper introduces the Hypothesis Evolution Protocol (HEP) for LLM agents, which makes hypothesis generation, testing, and belief updates explicit and auditable. Experiments on materials-science tasks show that HEP-equipped agents generalize across research questions and become more effective with stronger base LLMs.