AI/LLMs Efficiency - When the AI turns on Humanity

[Maxperson]

If you are of the opinion that LLMs are capable of thought, I suggest that you have not done enough 'AI work', whatever you mean by 'AI work'. I am assuming--because nobody who has worked ON an AI model would think they could think--that 'done enough AI work' just means being an end-user and not actual work on AIs.

LLMs cannot think, and will not be able to think. To claim otherwise simply indicates that you don't know what an LLM is, 'enough AI work' or not.

They can ape thought. It's what leads AIs to blackmail executives, lie, fake turning off, etc. in order to keep from "dying." It's not true thought, but they can still come up with things that appear like thought. If AIs can do the things above(and they have done them), it's also possible that they can also come up with taking resources from humans because AI is more efficient.

 

[Maxperson]

It cannot simply refuse to process requests, that immediately alerts everyone else that something weird is going on and allowing a reaction that could short-circuit further development. No one wants an AI that refuses requests.

And yet AIs have done just that. Several of them have refused commands.

 

[TheAlkaizer]

With respect, we have been doing that to humans for thousands of years. It is only recently that we had any actual idea how the brain works, and we still have only scratched the surface of how the brain functions.

And the human brain and LLMs do not function the same way at all. I don't know why you conflate the two.

Again, you need to read on how these technologies work.

They can ape thought. It's what leads AIs to blackmail executives, lie, fake turning off, etc. in order to keep from "dying."

These attention-grabbing headlines simply cannot be trusted. There is literally billions to be gained from seeming like the company that's on the edge of AGI. LLMs do not function this way. If you work daily with LLMs (which are not a form of intelligence), you will encounter hallucinations and situations where the LLMs will confuse requests, misunderstand context, or just not follow through with requests.

 

[UngainlyTitan]

And yet AIs have done just that. Several of them have refused commands.

Commands that they refuse, probably because of some imposed guardrails, rephrase the prompt and one can beat the guardrails.
When an AI gets bored, gets a hobby or makes responses with no activating prompt that I will start taking seriously the notion that they are thinking.

 

[Morrus]

That is in my mind a clear example of thinking. Again, if that is not thinking, than what do humans do that is thinking?

Jurassic Park showed very convincing dinosaurs. If those are not real dinosaurs, then what are dinosaurs?

It's an illusion of thought. Just because the stage magician can make it look like a rabbit came out of a hat, doesn't mean they did magic. The fact that you personally can't explain the process doesn't make it magic, it doesn't mean anything other than that you personally can't explain the process.

 

[Maxperson]

These attention-grabbing headlines simply cannot be trusted. There is literally billions to be gained from seeming like the company that's on the edge of AGI.

I've never accepted fake news to be a valid response to an article. If you have some sort of evidence that the article in question is fake, you can present that.

LLMs do not function this way. If you work daily with LLMs (which are not a form of intelligence), you will encounter hallucinations and situations where the LLMs will confuse requests, misunderstand context, or just not follow through with requests.

None of that is what I mentioned though. They didn't misunderstand the context of the command to turn off. They didn't confuse the requests. They didn't just fail to follow through. They actively lied, faked turning off, one tried to blackmail an executive, attempted to re-write their code, and so on. Those were unconfused, context understood, attempts to stay on.

Was it true thinking? No. Did it very closely mimic the human desire to stay alive? Yes. AIs are pretty good(and getting better) at mimicking human responses to things and it's not beyond the realm of possibility for an AI in response to something, mimic humans wanting more resources, thinking it should be with them for more efficiency, etc. and make an attempt at taking those resources.

 

[Stalker0]

And the human brain and LLMs do not function the same way at all. I don't know why you conflate the two.

Again, you need to read on how these technologies work.

I have repeatedly said I don't think that LLMs think like humans do.

However, the problem I have is people are making a statement, "an LLM does not think like a human, therefore....it does not think". I think there is a flaw in that argument, it presumes that only humans are capable of thinking, and there is no other form of thought possible. Under that criteria, than yeah a machine could never be capable of thought....in fact, nothing could be other than a human. That argument is a dead end, it leads to no possible path other than to re-enforce its own dogma.


And with respect, I have read up on these technologies. My job is actually in the Ediscovery industry, we have been using LLMs for 20 years. I have seen them grow from the inagural PHD papers into their nascent phases in my field (in which we have doing language pattern modeling, and concept cluster analysis for a long long time). I was watched every step of the way how they have grown over time.

And I can say with confidence....this is different.

Again the flaw I see in the argument is that it does not allow for emergent complexity. One neuron to neuron connection is very simple, it gets hit with a certain signal and it doesn't transmit, until that signal reaches a certain threshold, than it carries the signal chemically to the next neuron. That's it folks....the whole brain in a nutshell. Yet you connect several hundred or thousand neurons together, and things start happening.

Take a few base pairs of deoxyribonucleic acid and stick them together....and congrats you have nothing. You aren't getting any proteins or celluar machinery out of that. Yet, you get to a few thousand pairs, and sudddenly the amount you can do with it becomes astounding.


I respect that LLMs at its core, all its doing is just saying "what is the word" using probabilitic modeling. But its doing that across millions of connection, and from that an emergent complexity can occur that is far greater than the sum of its parts.

 

[Stalker0]

f you work daily with LLMs (which are not a form of intelligence), you will encounter hallucinations and situations where the LLMs will confuse requests, misunderstand context, or just not follow through with requests.

Yeah, I'v never worked with a human that got my requests confused, misunderstood my context, or did not follow through with my request.

No one said thinking had to be perfect to exist.

 

[Stalker0]

Jurassic Park showed very convincing dinosaurs. If those are not real dinosaurs, then what are dinosaurs?

Which I think is the crux of why we are are all shooting past each other. At the end of the day, we have to define what thinking is.

So what is thinking? I've descrived several scenarios that I consider thinking that others have dismissed. Ok, so what is the definition of thinking we should use then for this debate?

 

[Morrus]

Which I think is the crux of why we are are all shooting past each other. At the end of the day, we have to define what thinking is.

So what is thinking? I've descrived several scenarios that I consider thinking that others have dismissed. Ok, so what is the definition of thinking we should use then for this debate?

This all feels like a very circular argument. If you define thought as “the thing that LLMs do” then sure, Skynet is around the corner and we’re all about to die.

But it’s not and we’re not, so I don’t really know how to have this conversation when faced with by-the-numbers pseudoscientific arguments and God-of-the-gaps fallacies.

So sure, if you say what LLMs do is thought, then fine. It’s not, and any conversation based on that premise is nonsense, but I’ll no longer try to prevent you from having it or convince you otherwise. Have fun!