BeyondSWE is shaking up the benchmark scene for code agents, challenging them with tasks beyond single-repository fixes. With current scores revealing significant gaps, it's clear the journey is just beginning.
The EpiQAL benchmark aims to evaluate large language models' capabilities in epidemiological reasoning. Current LLMs struggle with multi-step inference, indicating a gap in their understanding of population-level health data.
A new PAC-Bayesian framework provides fresh insights into how physics-informed machine learning models generalize, particularly with unbounded losses. This approach offers practical benefits for training models with tighter bounds.
