[tag]
4 stories
A new arXiv study shows that multi-agent code judges often lack the specific evidence needed to distinguish between two solutions, leading to high rates of non-discrimination.
Researchers show that AI models fine-tuned on popular benchmarks inflate their scores by memorizing, not learning, and propose a framework to measure the real-world performance gap.
A research team proposes a rubric-based scoring system that makes AI agents’ self-assessments more interpretable and reliable.
A new evaluation framework probes how large language models behave in high-stakes medical settings, focusing on safety, calibration, and robustness to realistic clinical inputs.
