this is fucking cool. dude trained an ai model on his macbook by hacking into apple's m4 chip.

- he hacked into apple's inference accelerator (found in m4 chip) which turned it into a mini transformer.
- 80X more efficient than nvidia a100 gpus
- cost = electricity to charge macbook vs. renting gpus for $10Ks
- can be used to train, finetune and run models locally
my only feedback is its probably slow af but so cool regardless - you can imagine future chips (m6/7) can train much quicker.