The best AI Evaluation newsletters

Rapida's blog covers engineering and product lessons from building production voice AI, including how LLM latency metrics such as time to first token affect perceived responsiveness in live calls. It also shares an essay on why binary pass/fail evaluations with short commentary worked better than multi-criteria scoring for an AI coaching product.

The QURI blog, from a nonprofit researching forecasting and epistemics, announces its move into maintenance mode after about seven years and the founder's move to AI safety work at Google DeepMind. It also releases unpublished drafts and AI-assisted research wikis on delegation risk, reasoning processes and evaluation engineering.

This is the research and product announcement blog from PortexAI, covering open-source agent benchmarks such as COMPOSITE-STEM, the AsymmetryZero evaluation framework, and Harbor-based agentic evals. Readers get updates on how the company builds expert-written evaluations and integrates them with agent testing tools.