Tirthankar Ghosal

This work compares open-weight and API-served general-purpose, multimodal, and astronomy-specialized models using judge-based correctness and complementary reference metrics and highlights the role of domain-specific evaluation in determining which models, capabilities, and evaluation criteria are appropriate for scientific workflows.

Storage limitations, not compute power, dictate the scalability of whole-slide image embedding extraction, reshaping our approach to handling large-scale medical data.

Achieving unprecedented consistency in cross-representation learning, CoCoEvolve outperforms existing methods by leveraging one-to-one correspondences without extra annotations.

Mechanism recovery in AI-generated hypotheses falters in early reasoning stages, revealing vulnerabilities in scientific validity that could mislead experimental planning.

Verifying discoveries in autonomous science is now the bottleneck, overshadowing the ease of generating hypotheses.

Graph-native reinforcement learning can boost hypothesis generation in materials science by achieving up to 65% better traceability than traditional models.