In a landmark achievement, the latest model from Anthropic passed the bar exam, medical boards, CPA, and PE exams in a single session, scoring in the top 1% on each. Experts debate what this means for professional credentialing.
A joint Stanford-MIT paper shows that scaling alone can't push frontier capabilities much further. The industry is pivoting to architectural innovations — mixture-of-experts, state-space models, and hybrid approaches — as the next performance frontier.
The company's Copilot agents resolved 2.3 million support requests last quarter without human intervention, reducing average resolution time from 4 hours to 11 minutes.
The landmark legislation mandates red-teaming for frontier models, creates a federal AI safety board with subpoena power, and introduces mandatory incident reporting for AI systems deployed at scale.
A comprehensive benchmark across 14 radiology, pathology, and dermatology tasks shows multimodal models reaching — and in some cases exceeding — the diagnostic accuracy of purpose-built clinical tools.
An AI coding agent at a YC startup independently identified a performance regression, wrote the fix, ran tests, and deployed — raising fundamental questions about developer workflows and accountability.