Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

SYAI: Symposium AI and Data Challenges behind Emerging Self-Driving Laboratories

SYAI 1: AI and Data Challenges behind Emerging Self-Driving Laboratories

SYAI 1.5: Hauptvortrag

Donnerstag, 12. März 2026, 11:45–12:15, HSZ/AUDI

Transforming Our View on Transformers in the Sciences — •Kevin Maik Jablonka — Friedrich-Schiller Universität Jena, Jena, Germany — HPOLE Jena, Jena, Germany

The sciences face a fundamental challenge: much of the data we need for building models is not currently reported or usable in any structured form. Large language models (LLMs) have shown promise in addressing this through text-to-data conversion and direct predictive capabilities, attracting tremendous attention across scientific applications. In my contribution, I will present a roadmap based on comprehensive benchmarks of language models, multimodal models, and autonomous agents in chemistry and beyond. We have developed specialized evaluation benchmarks to assess scientific knowledge, reasoning abilities, and autonomous tool use across frontier models. Our results reveal a striking paradox: these models can exceed the performance of experts with decades of experience on certain tasks, while simultaneously failing worse than novices on other seemingly basic problems. I will discuss what these benchmark findings mean for the future of LLMs in scientific research, which applications are ready for deployment, where critical gaps remain, and how the field should prioritize development efforts. Finally, I will present the advances in models, systems, and infrastructure we have developed to enable practical impact across the sciences.