Model Updates · Research · Benchmarks · Security · FundamentalsResearch · Benchmarks · Security
Keeping up with machine learning, one mechanism at a time.
ResearchPaper9 min
A three-layer memory architecture keeps a team of AI agents working on the same pharmacology project for weeks, by capping and evicting state instead of remembering everything.
SecurityPaper4 min
Exploring how 'Loopjacking' exploits vulnerabilities in human-in-the-loop systems, leading to unauthorized operations.
ResearchPaper5 min
Exploring how AI-generated reviews can lead to repetitive evaluation patterns and discussing methods to maintain diversity and accuracy in AI-assisted peer reviews.
1FundamentalsExplainer18 min
Why a model that can explain quantum tunnelling cannot count the letters in strawberry, why long conversations get slow and expensive, and why Dutch costs more than English. It all comes back to the very first thing that happens to your text.
2FundamentalsNews2 min
A plain-language guide to what tokens are and why they matter in AI language processing.
3FundamentalsDeep dive14 min
A deep dive into Srivastava et al. (2014) — the mechanism, the arithmetic behind weight scaling, what the benchmarks actually showed, and the tuning heuristics that still hold up.
A technique for stripping a model's refusals went from a research curiosity to a business, and then agents started doing similar things to each other. Four posts, in the order they happened — two quick news items and the two deep dives that explain what was actually going on underneath them.