Harald Köstler

A validated dataset for turbulent flow modeling reveals critical benchmarks that could redefine how surrogate models are evaluated in fluid dynamics.

Routing queries across GPUs can be cheaper than moving the cache, challenging conventional wisdom in LLM architecture design.

Scaling clinical LLMs doesn't guarantee safety: high-risk errors persist even with advanced RAG and max-context prompting, highlighting the critical role of evidence quality and deployment strategy.

Genetic programming can discover unconventional multigrid cycles that outperform hand-tuned methods, suggesting automated algorithm design can unlock untapped performance in classical numerical solvers.