4 ms·
Can you elaborate on this terse statement? What parameter size would you train to? What will such a model be able to do that GPT3 cannot, and vice versa? What a
by etaioinshrdlu 6y ago
Can you elaborate on this terse statement? What parameter size would you train to? What will such a model be able to do that GPT3 cannot, and vice versa? What are you picturing?
- p1esk 6y agoI mean reproduce and improve/expand the dataset instead of training models.
- AgentME 6y agoWhat are you going to do with the dataset if not train models on it?
- p1esk 6y agoThe point is: training GPT-3 is too expensive to be a hobby project, but creating a good dataset isn't.
- etaioinshrdlu 6y agoIt's not too expensive for a group of hobbyists!
- p1esk 6y agoIt costs ~$4.5M to train GPT-3. How big is your group of hobbyists?