https://www.youtube.com/watch?v=oFfVt3S51T4

1. zero entropy bits should be tabbed away
2. eliminate all the low entropy actions that you perform in editor
3. speculative edits
4. smaller models for low latency predictions
5. input is huge the output is small
   1. the more context a model has less options to choose from
6. prompts should be cache aware
7. humans can't look in too many directions
8. there is an intelligence that went into building algorithms but the algorightms are not intelligent :P
9. as AI improves humans need to do more validations and verifications
10. cursor trained a model to apply the changes to a file
    1. this is not as straight forward as it appears
11. you should do what is more natural to you and we will try to figure out how to make that work
12. depth of thoughts you convey inside - Lex Fridman