The archive
Guides and explainers on artificial intelligence, newest first. Pick a topic to narrow things down.
Anthropic says Opus 5.5 costs 40% less to run than Opus 5 and tops its internal alignment audit — while warning that benchmark margins no longer track real-world differences.
5 min read
OpenAI has laid out four priority areas and seven principles for independent safety assessments of its models. The document is detailed on scope and access, quiet on who pays and who publishes.
5 min read
Anthropic says WHO AFRO, CEPI and INRB are using Claude to compress daily Ebola situation reports from a full day to under an hour. The unglamorous parts are the most instructive.
5 min read
An OpenAI customer story claims Higgsfield AI can now ship new creative features in a day with one engineer using GPT-6 Astra. The details behind that claim are thin.
5 min read
OpenAI claims an internal model resolved the Navier–Stokes Millennium Prize problem and 100-plus open problems, and is convening an independent mathematics advisory group at IAS.
5 min read
V7 turns enterprise documents into a queryable graph of entities and cited evidence, then serves it to agents over MCP. The architecture is sound; the vendor's own numbers need a careful read.
6 min read
AWS has added xAI's Grok 4.6 to Amazon Bedrock, bringing a 500K-token context window, four reasoning effort levels and tiered pricing across two different Bedrock endpoints.
4 min read
AWS and Benchling describe a multi-tenant sandbox for agent-written code that refuses DNS resolution by default, scopes storage access per job, and wires exfiltration tests into CI.
6 min read
OpenAI published a proposal urging the United States to lead the creation of international technical standards for frontier AI, including for recursive self-improvement.
5 min read
OpenAI expanded its Academy with role-based learning paths for developers, leaders, educators and college students, adding assessments that award a course badge.
5 min read
AI features create new ways for personal data to leak, persist, or be misused. These are the core privacy practices teams should build in from the first prototype.
7 min read
A workshop recap from Ai2 and Providence Swedish names five obstacles to AI-driven discovery, from scientific taste to the risk of amplifying flawed methods.
5 min read