No Backpropagation: Transformers Demystified in 60 Seconds!

Key Takeaways

white Transformers don't use back propagation to adjust its eror they do they do they do you train use train Transformers using a huge back propagation and the attention Ms get the matrices the keep bra M valy matrices get trained uh there however when you want to fine tune them right like this this Ms come out and then you have human human way to to train them more right so like you you put in front of a lot of humans several respon and that humans select response one response two Etc and for that you can use reinforcement learning so it's called reinforcement learning with human feedback because basically a transform is trying to play a game and the game is what score the human gives it right uh so you can use you can use RL uh or you can use other more direct methods for back propagation but the idea is that is that you do use back propagation