Comment on NoProp: Training neural networks without back-propagation or forward-propagationparentComments−ActorNightly1yI think we need to start thinking about one shot training. I.e instead of context into LLM, you should be able to tell it a fact, and it will encode that fact into the updated weights.
Comments
I think we need to start thinking about one shot training. I.e instead of context into LLM, you should be able to tell it a fact, and it will encode that fact into the updated weights.