pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
Why it matters
If base checkpoints already carry instruction and reasoning behaviour, assumptions behind fine-tuningTaking a trained model and training it a bit more on your own examples so it gets better at one specific job.Full definition → and evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → need revising.
Key quotes
“web text, which used to make up like up to 85% of the train data in GPT uh 3, is now all the way down at 15%”
Varun Singh
“RL was mostly just a cherry on top, um shaping the, you know, flavor of the interactions more than conferring extra um knowledge or quality onto the base model itself”
Varun Singh
“it makes sense to view supervised learning as a way specifically to prepare the model for to build useful representations for for RL”
Varun Singh
“base models have kind of moved from general, uh, human knowledge and world priors to reasoning and agentic behavior priors”
