Artificial Intelligence 2020 - is there progress? - page 55

 
Реter Konow:

... this technology will change the world - for sure.

It already seems to have changedReTeg Konow. And now it needs isolation)

 
Реter Konow:

You reassure yourself that it's no big deal? If it were a delusional generator, I'd be the last to worry about it...

You think human thinking is unimaginable and you can't create a powerful AI? What's stopping you from building a "clear model" in a machine? Apparently, GPT-3 isn't just a neural network (why would it be...) and there's something else in there. Otherwise it wouldn't work like that.

And the fact that this technology will change the World is certain.

It's not just a delusion generator, it's a highly advanced delusion generator.

I think human thinking can be imitated, but they can't yet. In time, I think they will learn.

In order to 'build a 'clear model' in a machine, you have to learn how to imitate human thinking in a machine.

 
sibirqk:

It's not just a delusion generator, it's a highly advanced delusion generator.

I think human thinking can be imitated, but not yet. In time, I think they will.

In order to build a clear model in a machine, it is necessary to learn how to imitate human thinking in the machine.

Not necessarily all thinking at once, but some of its functions first. And they have succeeded. Further on, there will be more. It is a start, after all.

 
Реter Konow:

It doesn't have to be the whole mind at once, some of its functions first. And they have succeeded. Then there will be more. This is the beginning, after all.

Well, as the saying goes - blessed is he who believes.

 
These guys spent 1,000 petaflops of days training the net to " squeeze out" from 500gb of literature, wikipedia and all sorts of internet crap algorithms for human thinking. I think they can be mastered and implemented at a lower cost. That's their flaw...
 
Реter Konow:
These guys spent 1000 petaflops of days training the net to " squeeze out" from 500gb of literature, wikipedia and all sorts of internet crap algorithms for human thinking. I think they can be mastered and implemented at a lower cost. That's their flaw...

I absolutely agree with this point of yours.

Only not "algorithms of human thinking", but probabilistic connections between words in the material being tested.
 
sibirqk:

I absolutely agree with this point.

Only not "algorithms of human thinking", but probabilistic relations between words in tested material.
I advise you to have an unbiased look at the work of GPT-3 in the article translated from the English source on the previous page. Unrealistic things are described there. Maybe they lie, of course, but why? It doesn't fit into the "probabilistic relationships". That's the problem.
 
Sent a request to provide an API for testing. If I get access, this is where we'll test the features and limitations.
 
Реter Konow:
Sent a request to provide an API for testing. If I get access, that's where we'll test the features and limitations.

Sounds like a threat - armed with the GPT-3 you will write 5 times longer posts on the forum and 10 times more often.

 
sibirqk:

Sounds like a threat - armed with the GPT-3 you will write 5 times longer posts on the forum and 10 times more often.

Well, if it makes you realise that the AI train is almost gone and you're still standing on the platform, then why not...