logoalt Hacker News

reactordevyesterday at 6:11 PM1 replyview on HN

Turn down the temperature and you’ll see less “simpler” short cuts.


Replies

smokelyesterday at 7:51 PM

For the uninitiated: Interestingly, it is not advisable to take this to the extreme and set temperature to 0.

That would seem logical, as the results are then completely deterministic, but it turns out that a suboptimal token may result in a better answer in the long run. Also, allowing for a little bit of noise gives the model room to talk itself out of a suboptimal path.

show 1 reply