google search is dead and AI finished it off

[solarbird]

The attached image is my reproduction of an attempt to use Google Search to find the name of a country in Africa that starts with the letter K.

And no, Bard is not better – this is another person’s test posted to Mastodon:

Posted via Solarbird{y|z|yz}, Collected.

Comments

[numb3r_5ev3n]

Since it's clear that Google has basically failed us: what search engine would you recommend now?

[solarbird]

I gave up on Google ages ago in favour of DuckDuckGo. It's improved over time; I think it's still not as good as Google was at its best, but it's entirely reasonable and doesn't have this shit. At least, not yet.

[sistawendy]

Woooooow. The first vendor to back away from AI search wins.

[solarbird]

It's genuinely stunning how many corporate jackholes have bet corporate farms on this absolute turnip of a technology.

[arethinn]

Whatever AI they're using there has abysmal short term memory. "None of them start with a K, except this one that starts with a K"??

[solarbird]

Well, that's kind of the point, isn't it? The 'ai' has no intelligence at all. It's just a probability-based procedural language generator. There is no intelligence involved, just statistics.

It's been likened to lossy compression/decompression, actually, and that ... makes a fair deal of sense if you dig into it. Kinda is.

[arethinn]

Sure, but there's generating bullshit, and then there's generating bullshit that's at least self-consistent and looks convincing! You'd think Google of all companies would have money to throw at it.

[marahmarie]

The latest explainer for LLMs increasingly dumbing down is that they train on the internet. People publish other LLM's wrong answers to the internet (either honestly thinking all of their answer's are correct or else in being too ignorant to know at least some of their answers are wrong) so the LLMs are now training on other LLM's wrong answers, making them, if nothing else, more evenly wrong across the board.

(Like maybe the first wrong answer found online was from Bard but then ChaptGPT "sees" it, snatches it up and regurgitates it next and so on down the line.)

I'm glad, actually, that LLMs are getting dumber simply by feeding off of each other's answers and that no one (hopefully, "ever") knows how to fix it. LLMs combined with various technologies already in the works are simply going to destroy humanity as they've always threatened to do anyhow and yeah, I can wait.