Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> If, based on the training data, the most statistically likely series of words for a given prompt is the correct answer, it will give correct answers.

ChatGPT does not work this way. It wasn't trained to produce "statistically likely" output, it was trained for highly rated by humans output.



Not exactly. ChatGPT was absolutely trained to produce statistically likely output, it just had an extra training step added for human ratings. If they relied entirely on human ratings there would not have been sufficient data to train the model.


The last step is what matters. "Statistically likely" is very underdetermined anyway, answering everything with "e" is statistically likely.

(That's why original GPT3 is known for constantly ending up in infinite loops.)


"e" is not a likely response to anything. I think you are not understanding the type of statistics involved here.


GPT3 doesn't create "responses". Not till it's been trained to via RLHF.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: