RunRL (@runrl_com) header
RunRL (@runrl_com) profile picture

Making reward go up

Joined January 2025
Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)Photo by RunRL (@runrl_com)
new @runrl_com blog post: i pretrained a tiny transformer model on perfect tic tac toe moves and measured how much it affects RL compute requirements
2
1
5
918
I have been waiting for this to drop for months
guilty as charged: we're a rapper company
1
14
3,964
A cool technique from the RunRL research team: using PD controllers to balance multiobjective loss functions! Suppose you want to train a model to give short yet relevant answers. For a given minimum level of relevance, this lets us improve on the other reward terms much more!
1
3
11
1,204
if you've ever wanted to run rl there's never been a better time to do it than now
2
1
9
484
opus 4 generating the RunRL headquarters in tikz
1
6
696
I am afraid to report, RL works.
25
36
8
817
111,627