True Positive Weekly #151
The most important artificial intelligence and machine learning news and articles
Hey, in this issue: Superintelligence and law; Large-scale online deanonymization with LLMs; The new reality of software delivery; Agentic workflows for software development; A practical guide to production-grade observability in multi-agent AI systems; Frontier model training methodologies; and more.
[Google] Teaching LLMs to reason like Bayesians
[Ars Technica] Large genome model: Open source AI trained on trillions of bases
[Google] Teaching AI to read a map
[Project] A tiny transformer (<100 params) can add two 10-digit numbers to 100% accuracy
[Project] Micro diffusion: Discrete text diffusion in ~150 lines of pure Python
[Tool] Monty: a minimal, secure Python interpreter for running code written by AI agents
This is it for this week. Please like, comment, share, or become a paid subscriber to support my work.