2 ms·
1. In-context learning - you can teach the model things without retraining the underlying model parameters. 2. Actual model updates (it's not like we'll be stu
by jstx1 3y ago
1. In-context learning - you can teach the model things without retraining the underlying model parameters.
2. Actual model updates (it's not like we'll be stuck with the 2023 version forever).
- roflyear 3y agoI don't think you can do 1, can you demonstrate that is possible? I don't argue you can't add to its trainingset, of course you can.