upvote
Well, LLMs handle raw data perhaps a million times slower than conventional algorithms, so in "bytes" and "latency" terms, a dialup modem can handle most AI communication today - a rack of cloud GPUs 100ms away is still faster for a lot of things than $10k of hardware in your GPU, Mac, etc.
reply