How LLMs and RLHF works | 3min snip from "Moment of Zen"

The AI moment, AI vs crypto: a heated debate with Amjad Masad, Flo Crivello, and Nathan Labenz

"Moment of Zen"

NOTE

How LLMs and RLHF works

Large language models, LMs for short, are trained through a process called pre-training where the model learns to predict the next word or token by analyzing the entire corpus of the internet. This pre-training step results in the emergence of intelligence in the model. Reinforcement learning from human feedback (RLHF) is a newer approach where an optimizing function is created to align the AI's predictions with human preferences. In RLHF, the AI is given instructions and if it follows those instructions, it is rewarded. If it deviates from the instructions, it is punished. This process is repeated to make the AI listen and learn from human feedback.

00:00

Transcript

Play full episode

Transcript

Episode notes

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.