Will AI get dumber?

[A Cardboard Box]

If your "field" involves animation and drawing that can be accomplished by an AI model then you are a waste of space.

Are you the pajeet "animator" who keeps creating improperly rigged models that wastes the time of valuable animators who have to get brought in to fix your mess?

Cope and seethe.

 

[SSj_Ness (Yiffed)]

If libtards keep lobotomizing it, yes.

 

[The Ugly One]

Enterprise AI is quite useful and only getting better as the companies that use it build more actual data. For example, inferencing on credit card fraud data to detect new frauds, or inferencing on inventory data to predict shortfalls and overstocks, etc.

 

[Fapcop]

Seems like google is sitting on a goldmine with decades of gmail data to use.

Cute that you don’t think they’ve probably already used it.

Pretty much in a nutshell, AI will never be permitted to get too smart.

I think you’re being optimistic here.

The thing is that AI has already surpassed us, in the sense that we sorta know how it works, but not exactly how or why.

We know that if we feed enough GB of input into an AI, whether Reddit posts or novels, it’ll deliver something that resembles intelligence.

How exactly it works we don’t know since the amount of data is just too huge and complex.

As an example: When you start feeding AI data from other AI, it invariably starts “hallucinating”. Why that happens? We don’t know. Just like numerous other examples of AI misfiring is shrugged off as “Welp, who knows why that happened?”

 

[byron]

Cute that you don’t think they’ve probably already used it.

Nothing I said implied that they haven't already used it.

 

[cybertoaster]

I guess that all those modifications to get AI to not say anything compromising or do any un-PC observations have fucked it up. Even since we crossed the 60 million parameter level AIs have become a sort of black box where data goes in and result goes out but we're not sure of what the entire process to get that result was, it "just works" most of the time, how the model transforms inputs into outputs—remain opaque. Convolutional layers make it harder for humans to understand what features the model is using for its predictions. Current models are on the trillions of parameters.

So for example sometimes when seeing code I've found comments saying "IDK what this does but when I delete it the entire program stops working". IMHO AI is a lot like this in that if you change something it changes other things. So by keeping the AI giving results that are "not convenient" they're messing up its entire reasoning, it can't say there are more than 2 genders without that fucking up its logic in other areas too.

That's my theory.

 

[Peepeepoopoo Witch]

AI are created, maintained by, and learn from humans, so they'll get dumber as humans get dumber. This is especially so if everything gets taken over by incompetent techjeets.

 

[Celso Bin Portiolli]

I saw an article about how AI images deteriorated as they started referencing other AI images.
I think the same thing might happen with information AI.
There is also the problem of companies not liking the results AI gives for not being politically correct and forcing them to give a different output, so that probably sabotages them as well.