@IsabellaMWeber@derJamesJackson Yup. I can confirm. Somehow there was a huge desire to capture the market yet be fully suspicious of any collaboration with local companies. Very weird contradiction not seen as such by most.
@neural_avb Cool write-up! Two questions:
1. If the context window was crashing, is there a "reward" (or penalty) for that? How did it give up crashing given that for weight update, you need valid outputs so gradient updates make this action less likely.
2. What is a "batch" in B_d?
@SteveLawrence_@resfoundation@annastansbury Care to elaborate why might that play such a crucial role? What skills / attitudes are learnt at that age that differ widely between kinds from affluent backgrounds and the poorer ones?