Alex was an applied scientist at Amazon and built distributed training systems during his time there. Given his experience, he methodically debunks several myths related to decentralized training.

Alex was an applied scientist at Amazon and built distributed training systems during his time there. Given his experience, he methodically debunks several myths related to decentralized training.
https://x.com/ai/status/1810332757807042738

i love how (some) ai and (most) zk researchers are using crypto narratives to fund their academic-esque research. pipemare, diloco et al are distributed, not decentralized. i believe we will get to promised land. 

not as soon as these projects/people imply and maybe not in the form they are selling.

thesis: decentralized ai will be on edge and on/near-device not on “cloud based blockchains” which you are tethered to like an umbilical cord

precedence/data for weakly held, strong opinion: this deep model could only run on a consumer/gaming nvidia gpu some years ago:

precedence/data for weakly held, strong opinion: this deep model could only run on a consumer/gaming nvidia gpu some years ago: https://store-usa.arduino.cc/products/gravity-huskylens-ai-machine-vision-sensor

fwiw, decentralized ai is already happening in the combat cloud. it’s a swarm env with trustless assumptions in some cases and getting more trustless by the day