6 ms·
> I’d say we got on the S-curve for “deep learning” about 20 years ago by that measure... You are correct that it's an S-curve. It's always an S-curve. It's
by AnimalMuppet 3y ago
> I’d say we got on the S-curve for “deep learning” about 20 years ago by that measure...
You are correct that it's an S-curve. It's always an S-curve. It's never an exponential forever. That said, parts of an S-curve can be close to exponential, for a substantial time.
> ... and we might be closer to the end than we are to the beginning.
For that to be the case, we'd have to be past the point of highest first derivative. My impression is that we aren't. We're either at it, or before it. (Hard to tell which, especially by gut feel, which is the only measure I've got, because hard numeric data on "GPT progress" is really scarce.)
We might be closer to the end than the beginning. I can't tell for sure. (That's how it goes, in the absence of hard data.) But my impression is that we're earlier than that.
- sitkack 3y agoYou also can't tell when you have stepped off of one S-curve onto the next one. We are on the next one that involves feedback and composition. This one is going to last awhile.