2 ms·
The thing that people strangely still seem to not get is that these systems will continue to improve rapidly. This baffles me as well. So many low hanging frui
by p1esk 2y ago
The thing that people strangely still seem to not get is that these systems will continue to improve rapidly.
This baffles me as well. So many low hanging fruits to pick, we literally just started with these large models. One obvious direction is predicting the next video frame. Infinite training data is available.
- qeternity 2y ago> One obvious direction is predicting the next video frame. This is being done. There are incredible generative video models...
- p1esk 2y agoI would not call them “incredible”. Video generation is still at gpt-1 level. It will need several orders of magnitude scaling to get to gpt-2 level. We’ve just started making very first baby steps in this direction.