The primer
Start at the beginning. Four short reads, in order, in plain English, each with a paper behind it for anyone who wants the detail. After them comes the more advanced part: how to use it, in six themes.
-
Primer 1 of 4What is an LLM?
What the model actually does, the four steps it is trained in, and the habit each step leaves behind. Open and closed models, and why the difference matters.
-
Primer 2 of 4What is a harness?
Everything built around the model: the six jobs it does, the everyday problems that are really harness problems, and why the same model scores so differently in different harnesses.
-
Primer 3 of 4What is context?
What the model can hold at once, how well it uses a long conversation, what the tools do when a conversation gets too long, and what anyone can do about it.
-
Primer 4 of 4What is “AI”?
The two together. What almost everyone means by “AI” right now is a model in a harness, and the rest of the vocabulary gets simple once you can tell them apart.
when Sam and Dario and the frontier vendors say "AI", theyre referring to an LLM (probability engines) wrapped in a harness of whatever flavor, there is no "AI model".
Berg Atkinson