935
I wish I was as bold as these authors.
(discuss.tchncs.de)
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 28 Jun 2024
935 points (98.7% liked)
Science Memes
9899 readers
2376 users here now
Welcome to c/science_memes @ Mander.xyz!
A place for majestic STEMLORD peacocking, as well as memes about the realities of working in a lab.
Rules
- Don't throw mud. Behave like an intellectual and remember the human.
- Keep it rooted (on topic).
- No spam.
- Infographics welcome, get schooled.
Sister Communities
Science and Research
Biology and Life Sciences
- !abiogenesis@mander.xyz
- !animal-behavior@mander.xyz
- !anthropology@mander.xyz
- !arachnology@mander.xyz
- !balconygardening@slrpnk.net
- !biodiversity@mander.xyz
- !biology@mander.xyz
- !biophysics@mander.xyz
- !botany@mander.xyz
- !ecology@mander.xyz
- !entomology@mander.xyz
- !fermentation@mander.xyz
- !herpetology@mander.xyz
- !houseplants@mander.xyz
- !medicine@mander.xyz
- !microscopy@mander.xyz
- !mycology@mander.xyz
- !nudibranchs@mander.xyz
- !nutrition@mander.xyz
- !palaeoecology@mander.xyz
- !palaeontology@mander.xyz
- !photosynthesis@mander.xyz
- !plantid@mander.xyz
- !plants@mander.xyz
- !reptiles and amphibians@mander.xyz
Physical Sciences
- !astronomy@mander.xyz
- !chemistry@mander.xyz
- !earthscience@mander.xyz
- !geography@mander.xyz
- !geospatial@mander.xyz
- !nuclear@mander.xyz
- !physics@mander.xyz
- !quantum-computing@mander.xyz
- !spectroscopy@mander.xyz
Humanities and Social Sciences
Practical and Applied Sciences
- !exercise-and sports-science@mander.xyz
- !gardening@mander.xyz
- !self sufficiency@mander.xyz
- !soilscience@slrpnk.net
- !terrariums@mander.xyz
- !timelapse@mander.xyz
Memes
Miscellaneous
founded 2 years ago
MODERATORS
Well, yeah. People are acting like language models are full fledged AI instead of just a parrot repeating stuff said online.
Spicy auto complete is a useful tool.
But these things are nothing more
The paper actually argues otherwise, though it's not fully settled on that conclusion, either.
Whenever any advance is made in AI, AI critics redefine AI so its not achieved yet according to their definition. Deep Blue Chess was an AI, an artificial intelligence. If you mean human or beyond level general intelligence, you're probably talking about AGI or ASI (general or super intelligence, respectively).
And the second comment about LLMs being parrots arises from a misunderstanding of how LLMs work. The early chatbots were actual parrots, saying prewritten sentences that they had either been preprogrammed with or got from their users. LLMs work differently, statistically predicting the next token (roughly equivalent to a word) based on all those that came before it, and parameters finetuned during training. Their temperature can be changed to give more or less predictable output, and as such, they have the potential for actually original output, unlike their parrot predecessors.
You completely missed the point. The point is people have been lead to believe LLM can do jobs that humans do because the output of LLMs sounds like the jobs people do, when in reality, speech is just one small part of these jobs. It turns, reasoning is a big part of these jobs, and LLMs simply don't reason.
That stems from the fact that AI is an ill-defined term that has no actual meaning. Before Google maps became popular, any route finding algorithm utilizing A* was considered "AI".
Bullshit. These people know exactly how LLMs work.
LLMs reproduce the form of language without any meaning being transmitted. That's called parroting.
AI is a marketing buzzword. When someone claims that so-called "AGI" is close, they're either doing marketing or falling for marketing.
Since you didn°t address the "parroting" part, I'm assuming that you retract your point.
Even if (and that's a big if) an AGI is going to be achieved at some point, there will be people calling it parroting by that definition. That's the Chinese room argument.
You're moving the goalposts.
Me? How can I move goalposts in a single sentence? We've had no previous conversation... And I'm not agreeing with the previous poster either..
By entering the discussion, you also engaged in the previops context. The discussion uas about LLMs being parrots.
And the argument was if there's meaning behind what they generate. That argument applies to AGIs too. It's a deeply debated philosophical question. What is meaning? Is our own thought pattern deterministic, and if it is, how do we know there's any meaning behind our own actions?
The burden of proof lies on the people making the claims about intelligence. "AI" pundits have supplied nothing but marketing-hype.
Which is what a parrot does.
Yeah this is the exact criticism. They recombine language pieces without really doing language. The end result looks like language, but it lacks any of the important characteristics of language such as meaning and intention.
If I say "Two plus two is four" I am communicating my belief about mathematics.
If an llm emits "two plus two is four" it is outputting a stochastically selected series of tokens linked by probabilities derived from training data. If the statement is true or false then that is accidental.
Hence, stochastic parrot.
If i train an LLM to do math, for the training data i generate
a+b=cstatements, never showing it the same one twice.It would be pointless for it to "memorize" every single question and answer it gets since it would never see that question again. The only way it would be able to generate correct answers would be if it gained a concept of what numbers are, and how the add operation operates on them to create a new number.
Rather than memorizing and parroting it would have to actually understand it in order to generate responses.
It's called generalization, it's why large amounts of data is required (if you show the same data again and again then memorizing becomes a viable strategy)
Seems like a pointless distinction, you were told it so you believe it to be the case? Why can't we say the LLM outputs what it believes is the correct answer? You're both just making some statement based on your prior experiences which may or may not be true
You're arguing against a position I didn't put forward. Also
This is what excessive reduction does to a mfer. That is just such a hysterically absurd take.
but, the LLM has faith!
I'm a curmudgeonly physics nerd, it's very strange being on the side of a debate going "No now come on, that's way too reductive"
That just means you’re better equipped when it comes up lmao
The AI builds some kind of a model of the world in order to better understand the input and improve the output prediction,
You have some mental model of how maths work which you have built up through school and other experiences in your life,
You both are given a maths problem, you both give an answer based on your understanding of mathematics
You’ve been speaking with your chest this whole time and now that we’re into the nitty gritty you really just said “The ai does... something!” It’s so general a description that by your measure automated thermostats are engaging in human reasoning when they make it a little bit cooler on a hot day.
You might’ve been oversimplifying on purpose. I just can’t help but think you have no idea how LLMs work outside of this inherently flawed comparison to human thought.
Not OP, but speaking from a fairly deep layman understanding of how LLMs work - all anyone really knows is that capabilities of fundamentally higher orders (like deception, which requires theory of mind) emerged by simply training larger networks. Since we don't have a great understanding of how our own intelligence emerges from our wetware, we're only guessing.
Something that looks like higher order reasoning emerged from training larger networks. At the end of the day it’s still just spicy autocomplete. Theoretically you could give it a large enough dataset to “predict” almost anything with really high accuracy, but all it’s doing is pattern recognition. One could argue that that’s all humans do, but that’s getting more into philosophy and skipping a lot of nuance.
I’m not like, trying to argue with you by the way. Just having a fun time with this line of thought ^^
What makes the "spicy autocomplete" perspective incomplete is also what makes LLMs work. The "Attention is All You Need" paper that introduced attention transformers describes a type of self-awareness necessary to predict the next word. In the process of writing the next word of an essay, it navigates a 22,000-dimensional semantic space, And the similarity to the way humans experience language is more than philosophical - the advancements in LLMs have sparked a bunch of new research in neurology.
The algorithm assigns weights to nodes in a neural network. These weights are derived by statistical association of tokens in the training data after they have been cleaned.
That is so enormously far from how we think humans learn (you don't teach a kid to understand theory of mind by plopping them in front of the Gutenberg project and saying good luck, and yet they learn to explain theory of mind problems all the same) that it is just comically farcial to assume something similar is happening underneath.
It is very interesting that llms are able to appear to be conversational but claiming they have some sort of mind with an understanding of maths is as ridiculous as suggesting a chess bot understands the Pauli exclusion principle because it never moves two pieces into the same physical space.
If you fine tune a LLM on math equations, odds are it won't actually learn how to reliably solve novel problems. Just the same as it won't become a subject matter expert on any topic, but it's a lot harder to write simple math that "looks, but is not, correct" than it is to waffle vaguely about a topic. The idea of a LLM creating a robust model of the semantics of the text it's trained on is, at face value, plausible; it just doesn't seem to actually happen in practice.
Prompt:
ChatGPT:
It's trained to generate what is most plausible, but with math, the only plausible response is the correct answer (assuming it has been trained on data where that has been the case)
ChatGPT uses auxiliary models to perform certain tasks like basic math and programming. Your explanation about plausibility is simply wrong.
It has access to a python interpreter and can use that to do math, but it shows you that this is happening, and it did not when i asked it.
I asked it to do another operation, this time specifying i wanted it to use an external tool, and it did
You have access to a dictionary, that doesn't prove you're incapable of spelling simple words on your own, like goddamn people what's with the hate boners for ai around here
That's not what I meant.
??? You just don't understand the difference between a LLM and a chat application using many different tools.
This is parrot libel
You take in some information, combine that with some precious experiences, and then output words
Which is what an LLM does.
Flat epistemological statements like this are why I feel like more STEM people need to take Philosophy.
Big fan of philosophy, so please do tell me how my joke is wrong? Does knowledge and beliefs not come from life experiences?
AI hasn't been redefined. For people familiar with the field it has always been a broad term meaning code that learns (and subdivided in many types of AI), and for people unfamiliar with the field it has always been a term synonymous with AGI. So when people in the former category put out a product and label it as AI, people in the latter category then run with it using their own definition.
For a long time ML had been the popular buzzword in tech and people outside the field didn't care about it. But then Google and OpenAI started calling ML and LLMs simply "AI" and that became the popular buzzword. And when everyone is talking about AI, and most people conflate that with AGI, the results are funny and scary at the same time.
Gamers screaming about the AI of bots/NPCs making them mad beg to differ
I was going to add a note about the exception of video games but decided I'm digressing
LLMs have more in common with chatbots than AI.
You are very skilled in the art of missing the point. LLMs can absolutely be used as chatbots, amongst other things. They are more advanced than their predecessors in this, and work in a different way. That does not stop them from being a form of artificial intelligence. Chatbots and AI are not mutually exclusive terms, the first is a subset of the second. And you may indeed be referring to AGI or ASI as AI, a misconception I addressed in my earlier comment.
I work on ML projects. I'm telling you, as a matter of fact, you do not understand what you are talking about.
Try being less smug and pedantic.
Oh, wow! You 'work in ML projects', do you?
Then maybe you could point out specific examples of me not knowing what I'm talking about, instead of general dismissiveness?
I'm not here to teach you and I don't care if you ever learn.
If you're interested check out your community college.
You have no obligation to teach me, correct. But if you choose not to, you have no right to criticise me without backing up your claims. Pick one.
I can absolutely criticize you without teaching you. No one is going to teach you 4 years of college and a decade of industry experience over a social media post so you stop lying online.
I appreciate you taking the time to clarify thank you!