Latest News

Why AI Judges Still Miss the Mark in Evaluating Research Agents

Current AI judges struggle with assessing complex research tasks, showing less than 55% accuracy. A new benchmark, REFLECT, aims to fix this.

MMoA: Rethinking Agent Selection in Language Models

Introducing MMoA, a new framework for language models that uses LSTM-based gating to enhance agent selection, reducing computational costs while maintaining accuracy.

The Language Gap in AI's Clinical Diagnosis: Why It Matters