Sunday: open-weights scale, AI in the lab, and a court ruling that sticks
Three threads worth sitting with this weekend: a new ceiling for open-weights models, an agent that found real drug candidates without a human in the loop, and a German court that just made AI-generated defamation a liability problem — not a hypothetical one.
Kimi K3 2.8T-A50B released: largest open model ever, Opus 4.8-class performance at Sonnet 5 pricing
A 2.8T-parameter open model at Sonnet-tier pricing compresses the cost-performance frontier in a way that changes what self-hosters can realistically run.
FutureHouse's Robin agent autonomously proposes drug repurposing candidates, validated in human cell experiments
Experimentally validated drug candidates from an autonomous agent — with human input limited to naming the disease — is a concrete benchmark for what agentic science pipelines can now deliver.
German Court Rules Google Liable for Defamatory AI Overview Search Results
Classifying AI Overview outputs as 'independent, substantive statements' strips the intermediary shield — a precedent that could reshape how AI-generated content is deployed across Europe.
Get the next edition in your inbox
Midweek and Sunday. Primary-sourced AI news paired with the guides that explain it.