The annual measurement volume: capability, investment, deployment, incidents, public opinion.
Why read it. Numbers, updated yearly, that let you check your own drift.
The clearinghouse
The best navigable map of the argument
Not a catalogue of scary AI stories, and not a reading list for optimists. Both cases are given the same seriousness, because a person deciding what to think needs the strongest version of each.
Long-form primary interviews where researchers and executives say more than they mean to.
Why read it. A high proportion of the quotable positions in this observatory originate here.
The closest thing the field has to an IPCC-style synthesis, commissioned after the Bletchley summit.
Why read it. Where the disagreements are documented rather than argued.
Separates predictive AI, which mostly does not work as advertised, from generative AI, which works differently than advertised.
Why read it. The best available inoculation against capability claims in press releases.
The first comprehensive statutory regime for AI, risk-tiered, with obligations phasing in through 2026 and 2027.
Why read it. Whatever anyone believes, this is now law, and it is the text that binds.
An extrapolation-driven argument that AGI by 2027 follows from straight lines on graphs, with national-security consequences.
Why read it. Whether or not the trendlines hold, this document shaped how a lot of capital and policy attention got allocated.
Technology raises general prosperity only when institutions force it to. Otherwise the gains concentrate.
Why read it. Reframes the AI question from 'how capable' to 'who decides', with a thousand years of evidence.
An early, contested claim that GPT-4 showed general-intelligence characteristics.
Why read it. A case study in the evaluation problem: the disagreement was about what counts, not about what the model did.
Argues containment of AI and synthetic biology is the defining problem, and that it may not be achievable.
Why read it. A frontier-lab leader making the containment argument from inside the industry.
Showed that loss falls predictably with compute, data and parameters, which turned capability into a budgeting exercise.
Why read it. The empirical basis for every timeline argument in either direction.
Seven hundred words arguing that general methods leveraging computation beat human-designed structure, every time, eventually.
Why read it. The shortest item in this library and among the most consequential.
The transformer architecture. Everything in the current debate runs on top of this paper.
Why read it. Provenance: the whole argument has a technical origin, and this is it.