Meta AI isn't there yet...

[Circlotron]

Take your pick... 

[artag]

No AIs are there yet. Even the simple one that does Youtube closed-captions is a complete disaster.
They're toys with severe limitations. 'Artificial' not in the sense that they aren't human, but, like artificial grass and artificial  flavouring, they're an imitation that replaces the real thing in only very restricted ways.

One day, I hope, we will have machine intelligences. LLMs might perform some low-power element of that, like an autonomous nervous system function or peripheral device,  But they're more likely a dead-end.

[Siwastaja]

It works as expected: generates words and characters one by one by picking the next one giving good probability that similar choice of words would occur in training text, generating output which looks plausible but is usually wrong, except by accident. Some people work in the exact same way, so you could really call it realistic artificial intelligence. Such people of course must be ignored in engineering contexts, but make e.g. great politicians.

[SiliconWizard]

In other words, that's noise. I don't think the world needs more noise, it's bad enough as it is.

[artag]

Such people of course must be ignored in engineering contexts, but make e.g. great politicians.

Successful politicians. Not great ones. Unlike in engineering, there's a world of difference.

[SiliconWizard]

So, AI isn't there yet, but the money is there already. So you may as well grab all you can while it lasts. Or something.

[tom66]

I've noticed Google has started integrating Gemini suggestions into its search results (at least for me).  Knowing how much compute generative AI requires, this must be costing them a fortune!  Most of the results have been accurate but some have been obviously wrong - I asked it about an obscure flag on a tar command and it just made up the answer. 

[Siwastaja]

I've noticed Google has started integrating Gemini suggestions into its search results (at least for me).  Knowing how much compute generative AI requires, this must be costing them a fortune!

This is unbelievable bullshit. Google's algorithms deteriorated day after day since 2004 or so. All they really need to do is to revert to original simple search algorithms and add some useful low-cost tuning; not deliberately break the algorithm for two freaking decades and then suddenly start jumping into another extreme end of complexity.

[SiliconWizard]

The search results deteriorated because the goal became to sell ads. I don't know how you can make a search engine, with such a business model, generally good for users. It just doesn't add up, unfortunately.

[tom66]

I've noticed Google has started integrating Gemini suggestions into its search results (at least for me).  Knowing how much compute generative AI requires, this must be costing them a fortune!

This is unbelievable bullshit. Google's algorithms deteriorated day after day since 2004 or so. All they really need to do is to revert to original simple search algorithms and add some useful low-cost tuning; not deliberately break the algorithm for two freaking decades and then suddenly start jumping into another extreme end of complexity.

Google's algorithms deteriorated because of the sheer amount of bullshit on the internet.  Entire jobs were created like 'SEO Optimiser' for instance, in an attempt to replace useful results with their result.  When the input to the algorithm is barely above the noise floor, there's no surprise that the output will often be noise. 

Conversely I am finding ChatGPT to often have the answer to technical problems I have, without having to sift through all the crap.   That's what Google is clearly aiming to compete with - I'm not sure they are there yet but they might get there eventually - trouble is, how do they monetise that?  AI trained to suggest products that advertisers pay for?