[archsurface]

I don't understand these instances of using ML to return search results. DDG, returns results; ML returns results, possibly with hallucination. Even without the hallucination what's the point? I find the results I need from a search engine. Solution looking for a problem?

[AlienRobot]

I don't know what Google is going to do with it, but Bing is absolutely ruining their search engine with AI. You search for something, AI starts typing out an answer in a typewriter effect. Since AI, I can't trust anymore that the snippets shown at the top are verbatim quotes from a human-written article or something the AI came up with (not considering they could be quoting an AI-generated article!). At the bottom of the search page, where the "next page" would normally be the rightmost button, it's now the "chat" GPT button. I misclick it every time I want to see the next page. I bet that is driving up some metrics and making some engineers really confused about why people keep clicking the chat button and then not chatting.

Overall this all feels so unimaginative. With all the resources these companies have the only solution they can come up with for the search problem is "just throw AI at it." I could come up with that. It's not clever.

[nolongerthere]

Curious why you ever use bing to begin with?

[AlienRobot]

They sponsor my browser of choice, Vivaldi.

Unlike Google, I can click the second tab every time and it goes to image search. Wait, actually they put "copilot" there and image search is the third tab now. Either way point stands: no shuffling of tabs.

Image search is actually better than Google. I can search for exact image sizes. Google used to offer this! I can just type my screen width and height and find the perfect wallpaper. Wait, it says "at least" here, not "exactly," so I guess it just stores the total amount of pixels of an image and then multiplies the width and height you inputted...

Can you believe this? It's 2024 and I can't even find an image by size on the Internet. I can't even trust the second tab is going to be the images tab. And some people think AI is going to fix software. It's ridiculous. It's just laughable. And so depressing.

[whimsicalism]

GPT4 is much faster for searching through results than I am typically and can pull out exactly what I need.

I've basically completely replaced Google in my day-to-day unless i need to look up a specific location of something in the physical world or something that recently happened.

[nicetryguy]

> I've basically completely replaced Google in my day-to-day (for GPT4)

That's ...not good.

GPTx gets alot of surface topics right but when you delve into gritty specific details it will just start rambling like a straight jacket lunatic with the confidence of a used car salesman. The rubber meets the road when i try to compile code that uses libraries or functions that don't exist or it leads me to hallucinated imaginary github repos. I worry that this use of GPTx would be like getting water from lead pipes: it would seem fine on the day-to-day while my mind is slowly poisoned with nonsense and insanity.

Google has certainly taken a nosedive in result quality for sure the last few years but Kagi has been amazing for me lately.

[whimsicalism]

Your usage of 'GPTx' makes me think you are conflating chatgpt and GPT4. I find chatgpt useless, so hopefully that's not what you're talking about.

None of the things you are describing happen to me, especially if you do basic trust+verify which you should be doing for Google anyways.

[diputsmonro]

Could you explain the difference and how you use GPT4? If all you're doing is hitting the GPT4 api, I don't see how it's different.

And, of course, you wouldn't know that your mind is being poisoned with hallucinate half-truths. Maybe you can pick some out because of prior knowledge, but what about the ones you can't? What about the little things you learn that you don't deem important enough to verify, but then remember later without remembering that they snuck in through an untrusted source? That's precisely the danger - you can't accurately tell truth from fiction, and the stuff you already know isn't the stuff you're asking about (otherwise you wouldn't be asking)

[whimsicalism]

> Could you explain the difference and how you use GPT4? If all you're doing is hitting the GPT4 api, I don't see how it's different.

The difference is that the models are completely different? I don't really find that GPT-4 hallucinates all that frequently (only in very nitty gritty details rarely).

> And, of course, you wouldn't know that your mind is being poisoned with hallucinate half-truths

Okay, so it appears you have some non-falsifiable theory of mind that somehow renders Google better because my mind is being poisoned. Not sure what sort of appeal to objectivity I could use to demonstrate otherwise.

> the stuff you already know isn't the stuff you're asking about (otherwise you wouldn't be asking)

True for Google - less true for GPT4, who I ask to give me practice problems and worked solutions of various things I already know about to practice.