alt Hacker NewsThat's the point. For things like summarizing a webpage or letting the user ask questions about it, not that much computation is required.
An 8B Ollama model installed on a middle of the road MacBook can do this effortlessly today without whirring. In several years, it will probably be all laptops.
Replies
You can just look down thread at what people actually expect to do - certainly not (just) text summarization. And even for summarization, if you want it to work for any web page (history blog, cooking description, github project, math paper, quantum computing breakthrough), and you want it accurate, you will certainly need way more than Ollama 8B. Add local image processing (since huge amounts of content are not understandable or summarizable if you can't understand images used in the content), and you'll see that for a real 99% solution you need models that will not run locally even in very wild dreams.
Sure. Let's solve our memory crisis without triggering WW3 with China over Taiwan first, and maybe then we can talk about adding even more expensive silicon to increasingly expensive laptops.
But what you would want to summarize a page. If I'm reading a blog, that means that I want to read it, not just a condensed version that might miss the exact information I need for an insight or create something that was never there.