3 ms·

I’m working in a related area and I’m rather curious about this point. In what way is GPT-4 overfit? Does overfit in this context mean the conventional: validat

by pama 4y ago

I’m working in a related area and I’m rather curious about this point. In what way is GPT-4 overfit? Does overfit in this context mean the conventional: validation loss went up with additional training, or something special?

rafiki6 4y ago
More specifically validation loss is irrelevant when you can't even sample out of distribution anymore.