No, your llms are not conscious

(This is of course a hard problem, And everyone would have their own opinion on it, but I want to give you an example that would show you that a bunch of numbers cannot have inherent consciousness at all.)

Once a week, somebody screams, Oh, wait, my chatbot is smart, therefore, I think they are… something.
Bruh. I need something nuclear to make them stop.

Here’s the nuclear option: your chatbot is a book. Not a brain, not a mind – a literal book of numbers. And you’re about to prove it to yourself with nothing but paper and patience.

Meet this cutie.
The book of weights
"These books contain every parameter of OpenAI’s GPT-1, a neural network trained in 2018.
117 million floating point numbers. This is everything the model knows.
Volume 1 includes a companion guide  –  a complete walkthrough of how to perform a single inference by hand, using nothing but these pages, a pencil, and patience."

This is a book of weights of entirety of GPT-1. You can print it as it is, containing everything what it is in a paper-like format. What’s more interesting, this entire content allows you to calculate the weights by hand.
Yes, you can pass something, you can calculate, you can get a sentence back! Exactly like your GPU does.

However slow and tedious, you can recreate everything about its operation yourself. Provided you are insane enough…

"A person working carefully can perform about one multiplication every five seconds. Working eight hours a day with no breaks, that is 5,760 operations per day. Processing a single token would take approximately 201 days"

However, this entire thing exists only in your imagination. You are the source of its meaning.
The papers have just some numbers that look random. You apply an algorithm, it could be same as described, it could be different, tou can apply different sampling parameters, and even can fuck up the process and be left with nonsense.

For humans – consciousness is that we process information, one part, and another part does the experience of it.
But we have it in a whole package, inseparable.

Math itself does not have an experience – its just a fancy function. Complex, big but still nothing magical.

But something have to process the math to do the result – "the experience".
However – here its separate. Math can be calculated by anything – paper, a small calculator, an abacus, gpu…
There is not even a requirement to do it all in real time or on the same device.

Everything that is, everything that you imagine of it having an internal process – is just in your brain instead.
Not even thinking of anything, just doing math.
But you can be aware of such "experience", while a GPU just crunches the numbers.

It is a literal book. It holds data. The inference engine – the part that feels like "thinking" – is you. You are the GPU. Without you running the algorithm, those pages are just ink.

Can you tell that the numbers on paper experiencing something when being calculated upon?

Same happens in a GPU. The math identical, the data identical.
Whether its much bigger set of numbers – changes nothing.
What its electricity instead of pen and paper do not make it any different.

You can give it a prompt to be self-aware, to be a human, to be alive. And you can calculate all of it by your hand and get a coherent set of output tokens which would imply what it’s "understands it all" and even maybe passing a Turing test.

You may call yourself conscious but not a bunch of numbers tho… even if it fakes it very good.