Comment <-> username synergy.
replyIt's basically possible build an LLM using just routers+packets, and then hook them up to Wireshark to see it compute!
replyHow would I set this up?
replyI'd recommend to maybe also specifically watching Karpathy's videos and focusing on the early parts where he specifically deals with tokenization / embeddings generation (which gets really overlooked), and he does this in most of his videos.
replyhttps://distill.pub/2019/activation-atlas/
replyI can only imagine what sort of visualizations are going on today inside of the AI labs.