Hacker News Reader

how do you deal with the fact that some basic pages can have tens of thousand of tokens?

Right now, not much. The extension is fairly basic in that is just looks at the raw text + HTML and sends it to the LLM.

The benefit of this approach is it's very simple and easy, but the downside is it sends a lot of unnecessary tokens to the LLM. That drives up the cost, slows things down, and hurts accuracy.

I'm working on a few improvements now to improve this.

I remember there was something called readability for chrome which is just what browsers have incorporated as reader view. And mozilla even had a stand-alone version of it (1). Might be of interest to you.

[1] https://github.com/mozilla/readability

Reply on news.ycombinator.com