Hacker News Reader

Some AI was done tastefully. Apple Photos search comes to mind. I can search for objects across your photos, and it does a reasonable job finding what I want. It's an example of AI that's so well done, the end user doesn't even know it's there.

Now Microsoft pushing "Copilot" is the complete opposite. It's so badly integrated with any standard workflow, it's disruptive in the worst of ways.

View on HN

Google's photo app had the searching for years. It seemed a lot more useful a few years ago when it was using dumber image recognition models. Now it returns fewer matches on the same search. I tend to search the same things a few times a year as a supplement to some story and it's been frustrating to see in real time, a picture that used to come up when searching "car" or "guitar" is missing and instead unrelated pictures returned.

That's fair, but my point was that AI should be implemented in a way that's out-of-the-way but is still helpful to users.

I think LLMs are incredible, I think there's a lot of really good usecases where it can help promote recommendations and actions for a user to take. I don't think every user wants to have every app they touch into a Chatbot though.

Reply on news.ycombinator.com