A 2-hour blackboard session watched at 1.25x speed
If you are like me and spend most of your time thinking about what happens inside the model,and not much on the hardware side of things this video will definitely fascinate you. Dwarkesh and Reiner Pope spent two hours at a blackboard going through the actual hardware economics of training and running LLMs and i got to learn a lot things i previously didn’tknow obviously. One of my biggest takeaways for me was the 6ND formula for calculating […]