Closing the RL loop. Instead of independent slop creators competing and TikTok determining the winners to show people, this turns the entire slop economy into an end-to-end differentiable monstrosity that we can point ADAM at and further hollow out the brains of young children.
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!
RunRL (@runrl_com) improves language models and agents with the power of reinforcement learning. Their customers have been able to go from 60% accuracy with Claude to 95% accuracy with RunRL.
https://t.co/kz5IC31tvi
Congrats on the launch, @dyushag and @dozenaltech!