GPT-3 and the Potential of AI-Generated Text - with OpenAI's Peter Welinder
The AI in Business Podcast
00:00
Is There Any Other Way to Train a Model Like This?
We trained it primarily on determining what word would come next in a string of text. It was somewhat controversial to crave a model like this t the time, because it's canof too simple. The first version of dptree was that simple. You needto be quite careful how you instruct the model to do something. For example, if you wanted the model to come with ta recipe, you sort of, like, what has mole been trained on is tonsor like recipes i found on the internet.
Transcript
Play full episode
