Andy
@gosalynboonchuy
ID: 858992998113136640
01-05-2017 10:34:24
11,11K Tweet
1,1K Followers
99 Following
Nando de Freitas LLMs donβt have communicative goals, they are just next token predictors. With RLHF they may bias the token distribution to maximize predicted human preferences, but they are not trying to explain or model the world by inventing abstractions. They just copy human abstractions.