A clear account.
An open trail of evidence.
LLM Primer helps you understand what changed in AI, why it matters, and how we know.
We select; we don’t collect everything.
We focus on developments that change how people understand, build, or evaluate AI. Models, methods, evaluations, and engineering are our starting point. An announcement is not important simply because it is widely repeated.
One development, one record.
Multiple accounts of the same development belong with one event. A new evaluation or a material follow-up may deserve its own linked entry. Dates describe what happened, not when someone happened to share a link.
A primary source is a starting point.
An official claim and an independent result are different kinds of evidence. We label author-reported results and explain what has not been checked. A schema check, a benchmark score, or an AI-written summary is not factual verification.
Uncertainty belongs beside the claim.
We name material limitations where you need them, including on share cards. Conceptual reading connections are distinguished from evidence of influence or causation. Historical context should not imply that later outcomes were predictable.
The record can change.
Material corrections explain what changed and when. Each entry identifies its revision and editorial status. Review editions contain source-backed working accounts awaiting publication review; they are not a complete daily news service.
Your time is the constraint.
A briefing should have an end. We show the actual review date, leave quiet periods quiet, and keep the reading public. There are no paid placements in this review edition.