Replying to
@protonprivacy@mastodon.social
I honestly think AI models should only be run locally. Many smaller models like the 7b variants can be run on cpu with 8gb of ram. These models may hallucinate information more often, but they are more than good enough for a lot of "productivity" tasks like writing emails, summarizing pages, etc
With ollama and the page assist browser extension, I would argue many devices used today can run decent LLMs locally (especially given CPUs tend to be overspecced for how they are used).