I struggle with the practicality of the whole thing.
The amount of tokens required to properly distill a frontier model is so large that by the time you could consume the # of tokens you would either be banned for extremely obvious abuse or a new model would be released, rendering your efforts less and less valuable over time. Intelligence is not a linear thing. Being behind just a little bit can have exponential consequences.
> Being behind just a little bit can have exponential consequences.
That seems to be the argument of Dario, Sam et. al., but I'm not ready to believe it. Time will tell, but this can be a marathon and Anthropic and OpenAI is in getting ready to sprint the last lap of the first mile.
I'm uneducated on how distillation works at more than a basic level so forgive me if this is a stupid question.
Isn't "distillation" of another provider's model exactly how these models got training date in the first place: Massive amounts of the written word + Prompt -> Answer. Why wouldn't distillation produce similar "reasoning" in the new model? It's just inputs and outputs.
What you're describing is (pre-)training. Distillation requires richer labels, the probability distribution over tokens (it would be logits rather than probabilities but that's not important). From a chat transcript you can only understand the argmax/most likely token of that distribution (and only if the API allows you to set the temperature to 0). It's not impossible for an API to give you that but they won't if they don't want you distilling their models.
The intuition is that distillation exploits not only the "right" answer but the relationship between answers (what's the second most right answer? the third? etc).
Among other things, because you simply can't get those "massive amounts" of text from a SOTA model at reasonable cost. And complex reasoning cannot possibly be trained in a pure one-shot fashion, real post-training takes massive resources. The whole story doesn't add up.
This is totally inaccurate, the APIs provide the reasoning logs. You ABSOLUTELY can distill from APIs, in fact, that's the primary way distillation is done currently.
The amount of tokens required to properly distill a frontier model is so large that by the time you could consume the # of tokens you would either be banned for extremely obvious abuse or a new model would be released, rendering your efforts less and less valuable over time. Intelligence is not a linear thing. Being behind just a little bit can have exponential consequences.