logoalt Hacker News

orangemoonxyesterday at 10:12 PM0 repliesview on HN

The prediction happens when it is uncompressed right?

LLM embeddings are compressed training data.

To decompress that is to make a prediction (in this case to convert the embedding into readable text)