Technology

hideshow 21 child comments

[–] Fizz@lemmy.nz 29 points 3 months ago (18 children)

How could that even be true since it was humans thay trained ai. What a dumb statement ahh

hideshow 18 child comments

parent

[–] communist@lemmy.frozeninferno.xyz 11 points 3 months ago (13 children)

Ai can do plenty of classification tasks better than humans though. It's not like every entity that trains another is fundamentally smarter.

hideshow 13 child comments

parent

[–] Fizz@lemmy.nz 7 points 3 months ago (5 children)

Yeah but to say humans can't do this when humans doing it were what provided it the data is a dumb statement. Humans classified the data so that AI could learn you wouldnt say humans can't classify data.

hideshow 5 child comments

parent

[–] communist@lemmy.frozeninferno.xyz -3 points 3 months ago (4 children)

Humans can, ai can do it better than humans in many areas though.

hideshow 4 child comments

parent

[–] Passerby6497@lemmy.world 4 points 3 months ago (1 child)

That's a MUCH different statement though, and not at all relevant to a claim that AI is able to do things humans CAN'T do.

parent

[–] communist@lemmy.frozeninferno.xyz 1 point 3 months ago

Protein folding, then.

parent

[–] Rothe@piefed.social -1 points 3 months ago (1 child)

In most areas it can't. It can do a much worse job though, if that is what you are looking for.

parent

[–] communist@lemmy.frozeninferno.xyz 1 point 3 months ago

So? In many it can do a much better one: protein folding for example

parent

[–] bold_atlas@lemmy.world 8 points 3 months ago* (6 children)

Software can't tell truth from lies because software doesn't even know what those are. It doesn't know anything. It's a parrot, a magic trick, a mirror to fool people (people like George Lucus) into thinking there's another person looking a back at them.

hideshow 6 child comments

parent

[–] communist@lemmy.frozeninferno.xyz 4 points 3 months ago* (4 children)

Regardless many ai classifiers outperform humans at classifying. Your comment seems irrelevant to the discussion.

hideshow 4 child comments

parent

[–] Rothe@piefed.social 2 points 3 months ago (1 child)

The parameters set for classifying that are basically non-existant. So no, they can't really do any of that. AI-bros will readily claim they can though, because it is like really important to them.

parent

[–] communist@lemmy.frozeninferno.xyz 1 point 3 months ago* (last edited 3 months ago)

I did not claim they can tell truth from fiction, it's just irrelevant. Goalpost shift for no reason on their part, they don't have to be able to do that to classify things or predict protein folding.

parent

[–] Passerby6497@lemmy.world 2 points 3 months ago (1 child)

Your comment seems irrelevant to the discussion.

As opposed to your comment moving the goalposts from an AI doing something humans can't, to them just being better than humans (which is absolutely arguable in most cases)?

parent

[–] communist@lemmy.frozeninferno.xyz 1 point 3 months ago

I was responding to the claim that ai's can't do anything better than the thing that trained them. It's completely relevant and not a goalpost shift. They can do plenty of things that humans can't feasibly do, such as protein folding prediction.

parent

[–] shuffle3765@sh.itjust.works 1 point 3 months ago

Your brain is also software

parent

[–] shuffle3765@sh.itjust.works 1 point 3 months ago (1 child)

When I was in college I built a chess playing program that I couldn't beat. It wasn't very good, but it was much better than me. The trick is machine learning: teach it the rules, get it to play millions of games against itself and adjust its strategy after each game.

The AI can become better at a learnable skill than those who trained it because the AI can spend zillions of hours practicing

parent

[–] Fizz@lemmy.nz 0 points 3 months ago

Yes but thats not whats happening here. The base is created by using human data.

parent

[–] M137@lemmy.today -1 points 3 months ago (1 child)

"thay"

parent

[–] Fizz@lemmy.nz 2 points 3 months ago

Useful comment

parent

[–] brsrklf@jlai.lu 11 points 3 months ago (1 child)

AI detection by AI is not reliable. Lots of false negatives, and comparatively less but still too much reporting false positives to be reliable (and those are especially biased).

It's unlikely to get better too. If AI was able to reliably detect AI output, why not use that to correct itself to be less AI-like? And where is the seemingly infinite money thrown at in this struggle?

parent

[–] Passerby6497@lemmy.world 3 points 3 months ago

If AI was able to reliably detect AI output, why not use that to correct itself to be less AI-like?

You just described adversarial training, something I've been hearing about for at least a few years now.