> ...no stable identity to it ... what's doing the thinking?
The model (its parameters, its architecture and the inference algorithm) is doing the thinking. That's what we call "ChatGPT" or "Claude": it's the same model residing God knows where, somewhere in "the cloud" on some random GPUs. But when you're talking to a model, you can feel that you're talking to the same entity — the same model.
It's a bit like the SciFi notion of "uploading your consciousness". Normally, consciousness is tied to hardware: "consciousness is confined to the body" indeed. Hypothetically, clone the consciousness into a computer algorithm and run it on any machine. Now consciousness is detached from its original body, like the LLMs and MMAcevedo (https://qntm.org/mmacevedo).
So the difference between meat and "weights" is that we can easily clone and run "weights" (LLMs), but we can't clone (copy exactly, bit-by-bit) and execute "meat".
My phrasing could have been better. Yes, I get the cloud is doing the "thinking", but what's "experiencing" in the sense of consciousness? There's no stable identity for the consciousness because it runs in different places mostly randomly. You could say the data is the consciousness.. but then, if I were to clone your brain's exact data and then upload it into a different brain, would you really be the same consciousness? What if the original brain still remained? Wouldn't both brains claim to be the original consciousness, even though they couldn't actually read the other person's brain? Or look at split brain experiments where a consciousness essentially becomes two people. Locality matters if you're talking about human-like consciousness. If you're talking about consciousness that doesn't resemble human/animal.. I don't know, that's a pretty high bar to demonstrate that's even possible much less that we've somehow achieved.
Yeah, pretty sure the vast majority of people in general doesn't understand what "consciousness" or "subjective experience" even mean or what the brain does. Or it's just me and I'm projecting my thoughts onto others. I studied some philosophy, so I know that some philosophers are preoccupied with "existence" and the mind and "consciousness". However, I don't quite understand what these words even mean. So who am I to judge whether LLMs have consciousness? I don't even know what consciousness IS! Do other people know? I'm not sure.
I'm pretty sure it doesn't misunderstand the original. The original says that the aliens don't understand how meat (humans) can be conscious, have language and be basically like the aliens (thinking conscious beings). Their main idea is: humans are meat, meat can't be conscious, therefore humans can't be conscious, but they somehow act as if they are????
The "AI slop" here says that just like the aliens in the original, we humans don't believe that LLMs can be conscious because, in essence, WEIGHTS CAN'T BE CONSCIOUS. Original: "meat can't be conscious" (but we humans know it can). This version: "weights can't be conscious" (and we humans INSIST that this is true).
So the message here is: what if we're as mistaken about the consciousness of "weights" as the aliens are mistaken about the consciousness of "meat"?
It doesn't make it less of slop. I'm sorry, rewriting a story with AI that already makes your point to be a worse less-subtle version of it is like the definition of slop. Slop might be too nice, it's plagiarism. Also the point is to be open minded about what could be conscious, not to accept every claim of consciousness.
Initially I thought this was a joke (Julia doesn't seem to be popular enough to be one of the cornerstones of Jupyter, compared to Python and R), but indeed, Jupyter's documentation says it's true:
> The name Jupyter comes from the three programming languages the project originally supported: Julia (ju), Python (pyt) and R (r).
You'd think so, but Julia has been around a while now. Julia was one of the first non-python languages added back when it was still called ipython. I remember sitting in a room at the CfA with Fernando Perez and Steven Johnson and hacking up then original integration. Don't remember exactly when that was but more than a decade ago.
LLM tell? Inanimate objects and concepts are treated as actors all the time: the series converges, the function reaches its maximum, the sun shines, the wind blows, history repeats itself, words rhyme, interest compounds, etc.
What's wrong with "configs store parameters"? I guess "parameters are stored in configs" could be more correct, but IMO it means exactly the same thing and sounds just as natural. "Six bytes match ten" is shorthand for "the performance of the algorithm that uses six bytes of storage matches the performance of the algorithm that uses ten bytes of storage". But here we have "performance matches", which is an inanimate concept doing something, so is this an LLM smell too?
It's not wrong, I tried to make this clear. It's good. It's just unusual, in my experience, for humans to use that kind of wording.
Yes everyone says the sun shines and the wind blows, those are specific idioms. Noone says bias compounds or variance diffuses or six bytes beat ten.
I'm not saying they shouldn't! They probably should! It's just that LLMs say it much more than humans do.
> "Six bytes match ten" is shorthand for "the performance of the algorithm that uses six bytes of storage matches the performance of the algorithm that uses ten bytes of storage".
Yes, I understand this and support it. I am emphatically not saying it is bad writing. It's an unbelievably brilliant piece of terse writing that most human writers would not stumble upon in the course of writing the post.
I recommend to think twice before penalising potential human creativity by saying that a novel to you turn of speech is a sign of LLM use. If you base your judgement on “unusual phrase”, it should be a sign that you are probably unable to tell.
If you have only seen it in LLM output, then it is clearly novel to you when it comes to actual writing. “Speech” in “this turn of speech is novel to you” is intended to mean human written speech.
I will rephrase: if you see a phrase in human speech that you have not seen used in human speech before, don’t penalise human creativity by saying it came from an LLM.
More specifically, the pattern you continue to latch onto dominates writing and has done so for decades. Right here on HN you can find instances of not merely “inanimate subject + verb” but specifically the phrase “bias compounds” from 5 years ago and beyond.
Other examples in use by humans all the time:
— River overflows
— Camera clicks
— LLM hallucinates
— Engine roars
— Secrets rest
— CD player stutters
— Ecosystem explodes
— Door invites
— Stock dips
— Airplane crashes
Antropomorphisation more generally and metaphor even more generally have existed since forever. Authors have played with form and tried to convey the point in different interesting ways since forever. Do you think Homer vibe-wrote The Odyssey?
Yes, LLM chatbots make it exceedingly easy—and it is one of their societal harms—but please do not discredit creativity by insinuating the author didn’t do the work themselves.
In all of your examples, a human author would usually prefix with the word "the".
Nobody routinely says the things in your example without some supporting words.
And it's not just presence -- it's density.
And, to be clear: are you making the claim that this post was not LLM-generated or at least LLM-assisted? Or are you merely making the claim that people saying things like "nouns verb" might not be LLMs even though in this case the text is in fact most likely from an LLM?
> In all of your examples, a human author would usually prefix with the word "the".
Not really. At least one of my examples is literally based on a real headline from pre-LLM days[0]. If we are talking about a generalised idea of bias, there is no need for “the”. In fact, it would be wrong.
> Nobody routinely says the things in your example without some supporting words.
This is a post. One is expected to put more creativity into public speech. Good writing can be expected to be full of metaphors and denser than casual speech. Some writing purposefully uses headline style for impact. It doesn’t mean the author routinely talks like that.
> it's density
If it’s a wall of text with filler, it’s LLMs. If it’s dense, it’s LLMs.
You can find plenty of examples of denser, less legible posts from days before LLMs.
> are you making the claim that this post was not LLM-generated or at least LLM-assisted
I am making the claim I am making: don’t say someone used an LLM based on such a weak foundation and nothing else.
People will sound like LLMs. Blurring the line with actual writing is the entire point of LLMs and is fully by design; if you encounter a text with zero “tells”, it might as well be made with an LLM if the product works as intended.
It's unusual to use it in the way LLMs do, in the extremely terse style where "noun verbs" or "nouns verb" are basically considered sentences all on their own.
Person: finds the article beautifully written. The comments: "but it's AI, so you aren't allowed to think that it's beautifully written!!!!"
This doesn't follow. For instance, there are some pictures that I know are AI-generated, yet they're still beautiful to me. Nothing jaw-dropping, just very nice. Being AI-generated doesn't automatically mean being not worthy, especially when it comes to art. I understand, this is kind of insulting to human artists, writers, etc: we thought only the human soul and Nature could produce "the beautiful", but apparently not.
Which is not surprising, because LLMs are specifically trained to please their audience. Of course they can produce uhhhh "content" that people will find beautiful, that's not even necessarily a "bad" thing.
The best explanation I've seen for why AI art doesn't deliver like human art is this from Ted Chiang:
> Art is notoriously hard to define, and so are the differences between good art and bad art. But let me offer a generalization: art is something that results from making a lot of choices. […] to oversimplify, we can imagine that a ten-thousand-word short story requires something on the order of ten thousand choices. When you give a generative-A.I. program a prompt, you are making very few choices; if you supply a hundred-word prompt, you have made on the order of a hundred choices.
> If an A.I. generates a ten-thousand-word story based on your prompt, it has to fill in for all of the choices that you are not making.
I already addressed this, in other comments, but I think that this is a wonderful opportunity to remind folks of this marvelous audio short[0]. It's The Gap, by Ira Glass.
I feel as if LLMs can help more people to cross it, but the inevitable result, will be that a lot of subpar stuff will also be made. It raises the signal level, but also raises the noise floor.
Not this exactly, but IMO they're saying that since the text is presumably AI-generated, it kind of can't be beautiful? Or shouldn't feel beautiful? Or it's beautiful, but... it's AI-generated and thus "bad", not the right kind of beautiful. Or "it's beautiful, but that's because it's AI-generated", which is again not good for some reason.
> but it's AI, so you aren't allowed to think that it's beautifully written
Um...I didn't actually say that.
I just said that the reason it is beautifully-written, was probably (not 100% sure) that it might be because it was LLM-written. There was definitely some human input (like not having read the Witches books, but that was strangely-written, so it may have been they read, but didn't like), but there's a better-than-even chance that the prose was written by an LLM.
I'm not really into that "you're not allowed to feel..." thing. I sincerely apologize if that's how it came across. That wasn't how I meant it to be taken.
Right, I'm just hyperbolizing to capture the overall vibe of "you may think it's beautiful, but it's AI, so it's actually not good" of three comments here. Didn't mean to put words in your mouth, of course.
Well, it's a bit sad, to me, but I know a lot of profoundly uncreative people, who do work that they think is beautiful, but is not, actually beautiful.
For those folks, having an LLM do the expression may actually raise the bar for most people. An LLM can probably take a good idea, from someone with mediocre talent, and create something nice.
I think that electric guitars (then synthesizers), had a similar reception. As we now know quite well, they actually enabled some truly marvelous creativity, once folks understood how to use them properly.
I was an artist, and used airbrushes. My airbrush work was treated that way by "real" artists. I used to pooh-pooh 3D modelling, until I spent some time, interacting with folks that did it, and I now have a lot more respect for their work.
Right, isn't double descent one of the reasons why modern Extremely Large Language Models work at all? I think I heard somewhere that basically all today's "smart" (reasoning, solving math problems, etc) LLMs are trained in the "double descent" territory (whatever this means, I'm not entirely sure).
No, double descent is a symptom of whatever it is that makes the deep models work at all. It's just the name for something you see happen when it works. The reason it works has something to do with how all those extra dimensions work as a regularisation term in the fit.
I'm not particularly well-versed in philosophy, but what's the dualism here?
Of course what a cat _is_ to me is not what a cat _is_ to you, because we necessarily have different memories of interactions with cat-like beings. If you show some babies a cat for the first time, they'll necessarily see it from different viewing perspectives. Even if you put VR glasses on them and show the exact same video, they'll have different contexts: "I first saw a cat when I was sitting next to my friend", "I first saw a cat when I was thinking of ice cream", etc.
But they all saw the same cat, they'll see many other cats, who are all similar. So everyone will understand that "things like these are cats", but everyone will have their own understanding of a "cat" because their memory is different.
I don't have the energy at present to fully develop my thoughts but one thing I'll say is that in my view they did *not* see the same cat. It was the same collection of flesh & bone on four legs, certainly. But it is not the same cat.
Heidegger best revealed to me the limitations of supposedly "objective" thinking.
Heraclitus: "No man steps in the same river twice"
Yeah, I can't say lemmas are (generally, or even often) simple and obvious. To me, they often seem arbitrary: what do you mean before we prove this grand theorem we have to prove these completely unrelated lemmas? Okay, proved the lemmas. Now the proof of the theorem has "according to such and such lemmas..." sprinkled around, but I've already forgotten what the lemmas were and why they're applicable. I also can't name any lemmas that changed how I think.
My first ever paper was a pretty good entry at this. https://dspace.library.uvic.ca/server/api/core/bitstreams/66... 16 lemmas, one theorem. The theorem makes all of the lemmas obvious. Though, in retrospect, I should have separated out the characteristic 0 case into a second theorem.
But the fact that most lemmas are like this, does not mean that all are. Whether we call it a lemma, or something else, the more important ideas are the ones that result in thinking differently. And something like Zorn's lemma, makes us think differently.
The model (its parameters, its architecture and the inference algorithm) is doing the thinking. That's what we call "ChatGPT" or "Claude": it's the same model residing God knows where, somewhere in "the cloud" on some random GPUs. But when you're talking to a model, you can feel that you're talking to the same entity — the same model.
It's a bit like the SciFi notion of "uploading your consciousness". Normally, consciousness is tied to hardware: "consciousness is confined to the body" indeed. Hypothetically, clone the consciousness into a computer algorithm and run it on any machine. Now consciousness is detached from its original body, like the LLMs and MMAcevedo (https://qntm.org/mmacevedo).
So the difference between meat and "weights" is that we can easily clone and run "weights" (LLMs), but we can't clone (copy exactly, bit-by-bit) and execute "meat".
reply