A benchmark with no peeking: double-blind AI evaluation
How a cryptographic enclave can test a closed model independently without exposing either its weights or the questions.
13 long-form stories · 10 source-led AI analyses
Breakthroughs worth checking.
In-depth bilingual analysis of recent AI research, with original visuals, transparent charts, primary sources and clearly stated limits.
Three useful perspectives
Choose the question, not just the date.
Start with a curated path or open the complete source-led library.
3 articles
When an agent crosses the line: four real incidents
What Anthropic’s disclosure teaches us about isolation, permissions and monitoring for long-running AI agents.
Can AI fix its own alignment failures?
Automated researchers found mitigations for ten measurable failure classes, while still requiring supervision themselves.
A map of nine billion possible DNA variants: AlphaGenome Atlas
A new atlas attempts to organize the effect of every possible single-letter change in the human genome.
WeatherNext 3: hourly forecasts on grids as fine as 5 km
A new model combines live satellite observations, frequent refreshes and high spatial resolution.
13 million lines of proof: AI formalizes Fermat’s theorem
Claude translated an established line of reasoning into a complete, computer-checked Lean proof.
The automated research intern: what changed inside the lab
Agents are taking on multi-hour research tasks, while planning, intervention and judgement remain human responsibilities.
Co-Scientist: a multi-agent partner for hypothesis generation
A Gemini system generates, critiques, compares and evolves hypotheses while the laboratory still decides what to test.
Did AI solve a Millennium Problem? What OpenAI announced
OpenAI published a claimed Navier–Stokes solution and a Lean formalization. Mathematical acceptance is only beginning.
GPT‑6 Astra is official: facts, limits and significance
Specifications, new agent mechanics, pricing and limits of OpenAI’s latest flagship model.
AI without the noise: five questions before automating
A practical way to separate useful automation from impressive features that only add complexity.
Designing technology people can trust
Trust is not a messaging layer. It grows from predictability, control and honestly communicated limits.
Curiosity as an operating system
How better questions, varied sources and small experiments create an advantage in a fast-changing world.