[ajs1998]

I want a browser to let me do more things, not fewer. I barely trust myself to press the right buttons on a page, I certainly don't trust an AI to buy shit for me. I don't have a million tabs open because I close them when I'm done, and I don't pretend to be good at multitasking.

AI browser integration offers no value to me. I find GitHub Copilot and similar coding agents to be useful only because they can index the codebase and documentation faster than I can. There is no analogous problem AI can solve for me in the browser unless websites dramatically change how they interact with requests from AI.

I like Zen. I'm sticking with Zen.

[felarof]

> There is no analogous problem AI can solve for me in the browser unless websites dramatically change how they interact with requests from AI.

Curious: Does summarizing, Q&A with web pages a problem for you?

[slipheen]

I am not the original poster, but I have a hard time understanding how this is useful to anyone.

I have tried these sort of tools before and never see the point – if it is something that I want to read in depth then I’m going to spend the time to read it.

If it’s something I’m not particularly interested in the details of, I’ll just skim it and read the parts that I /am/ interested in.

I am genuinely curious, can you help me understand why you would ever want to summarize a page?

This feels like skipping over your own life? Typically If I’m going to a webpage, it’s because I want to enjoy reading it.

[felarof]

I actually chat/Q&A with web pages quite often.

Examples:

- If I'm writing any email in gmail, I paste the enter email thread into chatGPT, claude and take its help with rephrasing or fixing grammatical mistakes

- I use chat to summarize HN threads too.

Definitely curious if the above use-cases are problems for you or not.

BTW when I want to do Q&A, I use LLM chat feature in BrowserOS -- which opens chatGPT, gemini, claude in split-view on any web page and I quickly copy paste the context using shortcuts.

[slipheen]

Thanks! I’m glad to see that that’s something that works for you, and I think I understand what you’re using it for more even if it’s not something I would do.

Personally - No way. I don’t want to let generative AI anywhere near anything I write in email.

If I’m writing an email to somebody it’s because I have something to say, and I want to express my own feelings on it.

(I suppose theoretically I could see it I guess if I needed to talk to someone who only used another language?)

But when reading/composing email, I’m not just trying to optimize for raw information throughput- I’m trying to experience my life and the lives of the people who took time to write to me.

If my friend wrote me a letter, I want to enjoy every word of it. Even for something more perfunctory, the nuance of the particulars in the way it is expressed carries a lot of information.

Likewise, I suppose with reading sites like HN comments- I want to see the perspective of the people who are writing, that’s the whole point of reading comments.

It helps me to understand them as people, their use cases, and how they interact with an particular article.

If I didn’t care about any of that, if I didn’t want their perspective, I wouldn’t bother reading the comments…

[ewoodrich]

Not OP but this kind of use case just isn't appealing to me personally. If anything, I'm delighted when I land on a website that's an old school wall of text with high information density and would rarely reach for a lossy LLM distillation built-in to the browser.

[ajs1998]

Not really. I could imagine AI-integrated docs pages being useful just from the huge volume of information available and how overwhelming it can be to learn something new.

I've used plenty of AI tools and models, local and paid, agents and chatbots, and the hallucination problem is a huge roadblock every time. I treat it like an autocomplete and I babysit its output because I have to. It just can't be trusted to give a correct answer.

I think LLMs being good at producing language is a trap. The chatbot interface is the obvious product to build around it, but it just sucks right now.