An interesting take, but we may be diverging in context.
Every token produced by an LLM is an extrapolation.
There are certainly models (like BERT) that have been specifically trained on interpolation, but GPTs are not.

That your argument was bad. And my point was correct.

You're trying to reframe your argument (which is good, because it was really bad) without invoking shit like NCAP (again, good)

I wasn't taking a dig at BYD- they're cool cars. I was taking a dig at your shit argument.
Like it or not, that was on-topic, because you made it part of the topic when you made the trash-tier argument.

No- it makes it a carnival ride. The gross simplification was on your part, in calling a carnival ride a "simulation", when in fact it's nothing even remotely close to one.

Our job here is to demonstrate the difference between a faithful simulation manifested in the world, and what you call the "real".
Demonstrating that an unfaithful simulation is not real is not hard.

Honestly, this is a trick question- and you won't win this debate. Minds have been trying to come up with a good answer for this for 2500 years with no success. You're not going to be the one that cracks that nut.

But once you realize that, you can liberate your mind from its biases that lead you to continue to come to illogical conclusions in the face of things that exist today in the real world.

That is how definitions work. Were you never taught how to use a fucking dictionary?
Words have multiple definitions, and if one of them is satisfied, the usage is correct.
If you claim that a word does not apply, and one of the definitions does apply- then you are wrong.
Calling it cherry picking is absurd- it's called "how to use a dictionary".

That is precisely how definitions work.
You are completely correct that it does not satisfy that definition of the word.
It however does satisfy another definition of the word, which means the usage is correct.

It's honestly amazing that I need to take you back to 4th grade English, but here we fucking go.
Let's apply your understanding of a dictionary.

To do this, we will try to determine if you're a person.
Now you might cherry pick 1a: human, individual, but since in your dumbshit universe we can pick any definition that doesn't match and preclude the word from correct usage, I point out that you most clearly are not 5a: one of the three modes of being in the Trinitarian Godhead.
Ergo, by your truly fucking IQ-of-65 reasoning abilities, you are not a human.

Congratulations, dipshit.
How in the fuck has someone with a UID that low survived so long being that fucking stupid? Are you maybe suffering from dementia?

Holy shit.

You're a fucking moron.
From the article you just linked:

Slow clap, my dim-witted friend. Slow clap.

I'm not offended in the slightest. Don't confuse colorful language and disdain for offense.
Evidence that thing A does action Z better than thing B is not evidence that thing A and thing B do not share properties. What kind of shit logic is that?

That's you trying to deflect the argument to a tu quoque.
I'm right because I'm right. Your fantasy of my emotions isn't relevant.

This isn't relevant to the discussion in the slightest.
I've been over this with you- an LLM does not have a concept of a query.
That you can put a query into its latent space and get a result is a feature of the network as a result of its training- but its context is not some kind of query space.

What an absurd claim.
LLMs take inputs into their context, and produce outputs.
They can (and are) quite literally used to operate in the real "analog" world.

No, I mean read.
Concepts like "understanding" are traps. You have defined them to be properties of some magical property you perceive your brain as having.
If input A -> FSM Y -> output B applies for 2 separate FSMs (And I invite you to demonstrate that your brain is not an FSM- but spoiler- you can't. Nobody ever has, and it's aphysical and unscientific to believe so) then we can satisfy any rational definition of "understanding".

So an illiterate person cannot think?

So can an LLM, once the image is processed by another model into embeddings that the LLM can understand (these are not words or tokens, because again, while the output layer of an LLM produces tokens, the network itself does not work with tokens- they are trained to turn their latent space into tokens)

Incorrect. You can write a word, draw an arrow to a thing, and the LLM will be able to understand that without problem. Try it.

No. Because our eyes are open, and we're not in denial.
Your argument is based on bad logic, and ignorance.
You claim LLMs cannot do things they demonstrably can do.
That is why you are wrong.

What I can't understand is the nature of that ignorance. All the tools are available to you to correct it. The denial appears willful.

It sure didn't.
Your dream of the human brain being more than classically physical, and ergo more than a finite state machine are not backed up by one single shred of evidence, and that's the Achilles Heel of all computers-can't-think arguments.
Maybe in another 1000 years you'll find that evidence, but it's not looking great for you guys.
Keep clinging to that notion of free will though.

I don't feel like I need to read an introductory book about my degree.
I'd be more than amused to see how you think it applies to the discussion at hand, however.
Looking for theories of computation to disprove thinking machines isn't going to get you far. I'll see your Daniel I. A. Cohen, and raise you an Alan Turing.

Yes, it is.

And your severe lack of a fucking clue made discussion with you an analogue to pissing in the wind anyway.
Name calling is just fun. You trying to use it to disengage is just you copping out.

No, I'm not.
There is no database, and the factors have nothing to do with words.

Fuck, you know so goddamn little about how these things actually work, and you just can't be assed to bother to fucking read, can you?

Humans misread shit all the fucking time, in any orientation.

Jesus fucking christ, you're dim.
You're acting like "Aha! These things can't read things at a 30 degree angle!!!", but that's an outright falsehood.
Is the error rate higher? Yes. Can they without error? Yes. It's all probabilities, just like humans.
If you've ever misread anything, I guess you've demonstrated that you can't think.

Google makes $350B a year.

You just said what it's done.
They have like 35 million paying customers pulling in ~$4B a year.
That's a far stretch from $52, but their growth is also obscene.

Of course the vast majority are free users. This is the case for all products like this. They'll hit $10B a year by next year unless there's a collapse in growth.

I think you probably shouldn't judge based on what you're seeing on the free shit models.
People are using it today in businesses. Agentic workflows are all over the fucking place.

Incorrect, in this case.
If we were talking about a restaurant in Epcot, or something like that- sure, I'd agree with you. But we're not.

We have AI to summarize the AI slop articles.

I'm the same, always use a good credit card, but the occasional hassle is worth it. The prices are 1/10th the Amazon ones.