RLHF, PPO & KL

Comment fonctionne GPT 3 ? Qu’est ce que “l’alignment” ?

Screenshot 2023-07-18 at 16.44.25.png

Screenshot 2023-07-18 at 16.56.46.png

Screenshot 2023-07-18 at 17.01.43.png

Screenshot 2023-07-18 at 17.41.53.png

Screenshot 2023-07-18 at 17.48.40.png

Screenshot 2023-07-18 at 17.44.45.png

Screenshot 2023-07-18 at 17.46.56.png

Screenshot 2023-07-18 at 18.00.27.png

Screenshot 2023-07-18 at 18.23.06.png

Screenshot 2023-07-18 at 18.31.31.png

Screenshot 2023-07-18 at 18.32.29.png