4 ms·
Wolfram pulled that out of a hat in an interview that was just a week or two ago. The data used for training GPT-4 is from before September 2021. The point Ste
by jiggawatts 3y ago
Wolfram pulled that out of a hat in an interview that was just a week or two ago. The data used for training GPT-4 is from before September 2021.
The point Stephen was trying to make was not about any specific sentence.
The point is that while forcing these models to get better through gradient descent, their only option for "going downhill" and improving the loss function is to go above and beyond mere grammar. That's because syntax and grammar only take them so far, and the only available source of improvement is to gain a general-purpose understanding of the world that the text they're seeing is describing.
- croes 3y agoI don't know if and how fast GPT learns from user input.
- pixl97 3y agoIt doesn't learn directly from user input. Instead there are two options. Taking the user input and putting it in the training corpus and reweighting the neural net. Or, using the user input as up/down votes on the RLHF to alter the output of the weights that already exist.
- dragonwriter 3y ago> It doesn't learn directly from user input. Depending on what “it” is, it does through in-context learning, though that’s, obviously, limited to the context window.