logoalt Hacker News

ByteAtATime • yesterday at 9:39 PM • 2 replies • view on HN

Interesting - I wonder if it's because coding doesn't use the specific token probabilities, while decision models do


Replies

anewhnaccount2 • yesterday at 10:23 PM

Yes this is the reason. Transformations like quantisation preserve the rank of outcomes much better than probability mass.

RussianCow • yesterday at 9:58 PM

I think it also helps that speed and latency aren't as vital for coding and we can afford to let the model think for longer.