@inthehands@hachyderm.io @Viss@mastodon.social A few years ago some Google engineer wrote a memo, “there is no moat”: https://semianalysis.com/2023/05/04/google-we-have-no-moat-and-neither/ Basically: small models are pretty good, and only going to get better — between OpenAI, Anthropic, and Google, and the open models the gap is small (I’ve seen the estimate that where OpenAI was six months ago is where the open models are now). That has persisted for the past three years since that essay was written. Lots and lots of people run local models on beefy desktops and even on laptops. Companies that are concerned about leaks of their IP use local models. It’s trivial to create an “agentic framework” that uses the local model LLMs as the LLM component, and those models are quite capable (for those applications where generative AI is useful at all). I think the data-center construction hype is a shell-game to con sovereign wealth funds out of billions, and to create an illusion of a “moat”.