HackerTube 中文?

The spelled-out intro to language modeling: building makemore

Andrej KarpathyYouTube ↗

Further reading 1

How far is next-token prediction from answering a person? →Separate three questions: how an input becomes a representation, how training changes a network and how a capable text predictor becomes an assistant. An explanation at one level does not settle the others.This video helps answer · What is the next unit, and how far back can prediction look?

Caption notes

Karpathy trains a character bigram model on names. It conditions on the previous character and can be evaluated from the probabilities it assigned to the sequence that actually occurred.

Public machine transcript (may contain transcription errors) ↗
Original passages · 2
we're always working with just two characters at a time
The spelled-out intro to language modeling: building makemore · ▶ 5:51–5:56
the product of all of these probabilities
The spelled-out intro to language modeling: building makemore · ▶ 53:21–53:26

Hacker News comments

View original on HN ↗