Models, research, and the systems reshaping how software gets built.
3 stories
A fresh round of independent benchmarking shows the best open-weight language models within striking distance of frontier closed models on reasoning tasks, narrowing a gap that used to be measured in years.
Devon AchterbergAI Correspondent
/2 min read
Lawmakers in several jurisdictions are advancing proposals that would require AI developers to disclose and document the provenance of training data, shifting regulatory focus from model outputs to model inputs.
Renata Cole//2 min
As AI agent deployments mature past demos, engineering teams report that retrieval quality, not model choice, has become the single biggest predictor of whether an agent actually works in production.
Devon Achterberg//2 min