hi!!

[Half Fast]

In machine language it’s compare and branch as I said in my first post. But it’s still an if then.

Even LLMs run, in the end, on digital hardware that uses Boolean logic at its core.

Yes, but that's reductive and has nothing to do with how AI actually works. Nobody talks about IF/THEN statements at a transistor level. At that point it's TRUE or FALSE and we're talking about individual bits.

You guys are arguing theory vs. practice. Yes, in theory, LLMs and AI and neural networks are non-Boolean. In actual practice, the only practical way to implement them at the moment is using a binary computer. If we want to implement true AI, we may need to return to room-filling analog computers and allow some noise on the signals.




Actually, research is in progress to develop modern analog computers with an eye toward using them for deep learning.

New hardware offers faster computation for artificial intelligence, with much less energy

MIT researchers created protonic programmable resistors — building blocks of analog deep learning systems — that can process data 1 million times faster than synapses in the human brain. These ultrafast, low-energy resistors could enable analog deep learning systems that can train new and more...

news.mit.edu

 

[TCABM]

@schmookeeg likely could give the best answer but he hasn’t been around in a while.

 

[sansoneservices]

It is highly-advanced automation and little more.

Correct ^^^ computers still respond only according to the data programed into the THING.. No intelligence yet, just a glorified file cabinet. Maybe someday

 

[Salty]

You guys are arguing theory vs. practice. Yes, in theory, LLMs and AI and neural networks are non-Boolean. In actual practice, the only practical way to implement them at the moment is using a binary computer. If we want to implement true AI, we may need to return to room-filling analog computers and allow some noise on the signals.

View attachment 135764


Actually, research is in progress to develop modern analog computers with an eye toward using them for deep learning.

New hardware offers faster computation for artificial intelligence, with much less energy

MIT researchers created protonic programmable resistors — building blocks of analog deep learning systems — that can process data 1 million times faster than synapses in the human brain. These ultrafast, low-energy resistors could enable analog deep learning systems that can train new and more...

news.mit.edu

I thought I made that clear in this post

All that fancy logic compiles down to if then statements in the end. No, it’s not coded that way, but inevitably that’s what it is.

 

[Half Fast]

I thought I made that clear in this post

You were perfectly clear about the method of implementation, but Baked Potato was focusing on the theoretical side and you guys were going right past one another.

 

[Stevel]

Blah blah blah. It makes no difference to me. I'm gonna cheat on Siri with Hannah and then brag about right here on this forum just to make all of you turn green with envy.

 

[William Pete Hodges]

Is Large Language Model anything like a Neural Network? It is my understanding that a neural network works in a similar manner to our own brain synapses. Is that true?