Utsav Dhanuka

Abhranil Chandra, Sankaran Vaidyanathan, Utsav Dhanuka et al.

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.

Utsav Dhanuka, Soham Poddar, Saptarshi Ghosh

In an era where public health is increasingly influenced by information shared on social media, combatting vaccine skepticism and misinformation has become a critical societal goal. Misleading narratives around vaccination have spread widely, creating barriers to achieving high immunisation rates and undermining trust in health recommendations. While efforts to detect misinformation have made significant progress, the generation of real time counter-arguments tailored to debunk such claims remains an insufficiently explored area. In this work, we explore the capabilities of LLMs to generate sound counter-argument rebuttals to vaccine misinformation. Building on prior research in misinformation debunking, we experiment with various prompting strategies and fine-tuning approaches to optimise counter-argument generation. Additionally, we train classifiers to categorise anti-vaccine tweets into multi-labeled categories such as concerns about vaccine efficacy, side effects, and political influences allowing for more context aware rebuttals. Our evaluation, conducted through human judgment, LLM based assessments, and automatic metrics, reveals strong alignment across these methods. Our findings demonstrate that integrating label descriptions and structured fine-tuning enhances counter-argument effectiveness, offering a promising approach for mitigating vaccine misinformation at scale.