I do everything pretty much these days at temperatures zero. I tend to use higher temperatures and then change the top P parameter, which is a sort of variation on this procedure that it trims the distribution of the long tail and then picks randomly from what's left. Usually the reason I'm doing some like that is some fighting against mode collapse, right? And there's cases where you want to be even more diverse than that. It becomes sort of an empirical problem of what maximizes performance.
This is a special preview episode of The Cognitive Revolution: How AI Changes Everything. Hosted by Erik Torenberg and Nathan Labenz, TCR hosts in-depth interviews with the creators, builders and thinkers pushing the bleeding edge of AI. On this episode, they talk with Riley Goodside, the first Staff Prompt Engineer at Scale AI and expert in prompting LLMs and integrating them into AI applications.
Check out The Cognitive Revolution The perfect AI interview complement to The AI Breakdown https://link.chtbl.com/TheCognitiveRevolution Find TCR on YouTube: https://www.youtube.com/@CognitiveRevolutionPodcast