As a software developer who took an elective in neural networks - when people call LLMs stochastic parrots, that's not criticism of their results.
It's literally a description of how they work.
The so-called training data is used to build a huge database of words and the probability of them fitting together.
Stochastic because the whole thing is statistics.
Parrot because the answer is just repeating the most probable word combinations from its training dataset.
Calling an LLM a stochastic parrot is lile calling a car a motorised vehicle with wheels. It doesn't say anything about cars being good or bad. It does, however, take away the magic. So if you feel a need to defend AI when you hear the term stochastic parrot, consider that you may have elevated them to a god-like status, and that's why you go on the defense when the magic is dispelled.
LR
Als Antwort auf Leeloo • • •Nuwagaba Gift
Als Antwort auf LR • • •How the parrots be justified?
Robin Palotai
Als Antwort auf Leeloo • • •Wolf480pl
Als Antwort auf Leeloo • • •webhat
Als Antwort auf Wolf480pl • • •Philip
Als Antwort auf Wolf480pl • • •Hypolite Petovan
Als Antwort auf Philip • • •Leeloo
Als Antwort auf Hypolite Petovan • • •@hypolite @wolf480pl @pkal
I've never seen cars advertised as freedom here.
Maybe with good reason, because when you are used to cycling, being stuffed into a metal box doesn't sound much like freedom.
And while most people have cars, their main reason is to get to work, which is not exactly freedom either.
Philip
Als Antwort auf Leeloo • • •Hypolite Petovan
Als Antwort auf Philip • • •Leeloo
Als Antwort auf Wolf480pl • • •Of course it can not be intelligent, it's just a huge database of probabilities.
Wolf480pl
Als Antwort auf Leeloo • • •pretty sure that's a fallacy, kinda like "a sculpture is just stone, therefore it can't be beautiful", or "a cell is just a bunch of proteins, therefore it cannot be a living creature".
Now, I'm not saying a huge database of probabilities can be intelligent (I hope it can't), just that I think a better argument is needed why in the case of a database of probabilities, what it's made of prevents it from being intelligent.
Leeloo
Als Antwort auf Wolf480pl • • •@wolf480pl
You would have to redefine intelligence for asking whether a list of numbers is intelligent to even make sense.
And your comparison is completely off. Beauty is not a property of the sculpture, it's, as they say, "in the eye pf the beholder". Some people find curves beautiful. Can a stone have curves? Yes, of course. Others may find sharp edges beautiful. Can a stone have sharp edges? Again, yes.
I suggest you consider once again whether you are elevating "AI" to a god-like status.
John Tinker
Als Antwort auf Wolf480pl • • •The effect that you are noticing is because the writers of the training material were intelligence. You are seeing the reflection of their intelligence in the output of the LLM: Here is output from an LLM that describes what an LLM is, and what it is not: johntinker.substack.com/p/misu…
Misunderstanding the Chatbot, seen as a Commutation Operator
John Tinker (John’s Substack)eestileib (she/hers)
Als Antwort auf Wolf480pl • • •Yes and I take that position.
Kay Ohtie
Als Antwort auf Leeloo • • •I hadn't thought about it as being something that takes magic away from folks like that. Honestly I always found it an accurate shortcut term for what's genuinely a fascinating but hilariously misused technology.
I think the worst part is then when folks hear "statistics" and go "See this is why it's safe to feed it raw data" and it's like oh my god NO.
calcifer
Als Antwort auf Kay Ohtie • • •@KayOhtie honestly it’s safe to feed a model pretty much anything
But where you direct the outputs and how they are acted upon can get incredibly dangerous amazingly quickly. There’s a common misbelief that if you’re careful about inputs, LLMs are safe; and that’s almost exactly backwards
Hypolite Petovan
Als Antwort auf calcifer • • •Kay Ohtie
Als Antwort auf calcifer • • •James Wood
Als Antwort auf Leeloo • • •Les Orchard
Als Antwort auf James Wood • • •James Wood
Als Antwort auf Les Orchard • • •Hypolite Petovan
Als Antwort auf James Wood • • •@James Wood @Les Orchard You realize that you literally prompted parroting, and it succeeded? What part of "stochastic parrot" did you think it was defeating?
Did you already offload some of your thinking to these systems?
calcifer
Als Antwort auf James Wood • • •@mudri @lmorchard it’s not inductive at all though. It’s just parroting the patterns it sees in its training data. If it wasn’t common to see exchanges like that, the response would be utter nonsense.
People misunderstand what “training” is. It’s modeling the input. Humans develop the rules for how to model that input. Emergent properties of that process can easily *seem* like thinking or reason, but it’s an illusion.
arclight
Als Antwort auf Les Orchard • • •@lmorchard @mudri Be careful not to conflate the actual language model with its user interface. Whatever was sent to or received from the LLM went through the chatbot layer. Or possibly was handled by thd chatbot layer without ever touching the LLM. We don't know because the whole system is opaque.
This casual experiment may not be telling you what you think it's telling you. :)
Mathias Hasselmann
Als Antwort auf James Wood • • •@mudri Because the prompt processor is explicitly programmed to recognized direct imperative commands containing words like "say", "repeat", "output", "print". Just like Eliza already did. You've got impressed by a programming technique from 1964. Congrats, Sherlock.
@leeloo
Growlph Ibex
Als Antwort auf Leeloo • • •calcifer
Als Antwort auf Growlph Ibex • • •@growlph this is the whole frustration I have with the polarization on the topic. There is genuinely utility. There’s also a very good argument that the utility doesn’t exceed the costs (socially, environmentally, etc).
But the hype is unreal and legitimately dangerous.
Tobias Ernst
Als Antwort auf Leeloo • •TAL mag das.
Leeloo
Als Antwort auf Tobias Ernst • • •Not with the current methods, and very lilely not without understanding a lot more about how pur own brains work.
James Baillie
Als Antwort auf Tobias Ernst • • •Frank Heijkamp
Als Antwort auf Tobias Ernst • • •A LLM is not able to reason. It can fool you into believing it is intelligent and self aware, where in fact it just parrots the patterns it has stored. These patterns are however very human-like as they are the result of training on texts written by actual humans.
The fun part starts now where the entire internet got flooded by #ai generated content. All of this will be the training set for the next generation of LLM's. What could possibly go wrong?
@leeloo
Hypolite Petovan
Als Antwort auf Tobias Ernst • • •@Tobias Ernst @Leeloo We are already way past that point, although it isn't distributed evenly. One of the reason is that LLMs are machine learning applications, and machine learning is extremely effective at reaching its stated goals, the problems being to define those goals, and that they are hidden as a trade secret by the major LLM companies.
But it isn't difficult to figure out that these companies favor outputs that looks and sounds as human as possible, in order to exploit our innate tendency to seek humanity in looks and sounds, including language.
Cluster Fcku
Als Antwort auf Leeloo • • •nina splendorr 🌻🏳️⚧️
Als Antwort auf Cluster Fcku • • •Gregory
Als Antwort auf Leeloo • • •I myself like calling LLMs "glorified autocomplete". Or "Т9 на максималках" in Russian.
It's surprising just how defensive some people get when I say that even when they agree with my definition. They keep believing that just give this thing more parameters and something magical, something more than sum of its parts will emerge, any moment now, just one more model generation, just one more order of magnitude, I promise.
Frank Heijkamp
Als Antwort auf Gregory • • •@grishka
The fun part is that the next generation will have the current state of the internet as its training set. An internet that is flooded by #ai generated content.
The biggest issue those ai companies face at the moment is how to only ingest human generated content and filter out as much as possible of all of the ai generated crap that is out there.
Good luck with that.
@leeloo
liffy 💜
Als Antwort auf Leeloo • • •Play Ball and Fight Fascists
Als Antwort auf Leeloo • • •usuario@instancia.org
Als Antwort auf Leeloo • • •Frank Heijkamp
Als Antwort auf usuario@instancia.org • • •I also had to look it up, I am however not a native speaker.
@leeloo @knuxbbs
International Journal of Matti
Als Antwort auf Leeloo • • •Tobias Ernst mag das.
foundseed
Als Antwort auf International Journal of Matti • • •Androcat
Als Antwort auf Leeloo • • •Sensitiver Inhalt
Uriel Fanelli
Als Antwort auf Leeloo • • •nope. What you describe as "stocastical parrot" is Markov, Hidden Markov Model (HMM) , not a VLLM.
You can find an HMM in your mobile phone, AKA T9, AKA "keyboard suggestions".
Leeloo
Als Antwort auf Uriel Fanelli • • •@uriel
What part exactly are you saying nope to.
Dispelling the magic and god-like status or some specific detail?
Uriel Fanelli
Als Antwort auf Leeloo • • •Uriel Fanelli
Als Antwort auf Leeloo • • •Hypolite Petovan
Als Antwort auf Uriel Fanelli • • •Troed Sångberg
Als Antwort auf Leeloo • • •A much better answer is "So are humans".
(according to everything we've so far been able to document regarding our own processes)
Leeloo
Als Antwort auf Troed Sångberg • • •@troed
The part that we understand about how our brain works is so simple that we can understand it.
The rest, we have no clue about.
Replicating the simple parts and pretending that will get us anywhere close to intelligence is the kind of magic I'm talking about.
Troed Sångberg
Als Antwort auf Leeloo • • •We don't know that. It's equally likely that we have a belief in that there must be some kind of "magic" in our brains that there simply isn't.
From a physics standpoint there can be no magic - the brain is just a large neural network with various inputs (wind blowing on arm hair etc) that results in outputs (mouth moving).
Resuna
Als Antwort auf Troed Sångberg • • •@troed
No, this is not just not true, it's absurdly not true.
Most of human thought isn't even language-based, let alone being representable as some kind of token generation. Most human thought is based on platforms that evolved long before language, that are demonstrably more capable than large language models at reasoning about the real world, since other entities that share these platforms are able to demonstrate quite sophisticated reasoning without involving language.
Troed Sångberg
Als Antwort auf Resuna • • •@resuna At no point am I stating that LLMs are exactly like human brains.
blog.troed.se/posts/the-delta-…
The delta between an LLM and consciousness
Things I couldn't find elsewhereHypolite Petovan
Als Antwort auf Troed Sångberg • • •Troed Sångberg
Als Antwort auf Hypolite Petovan • • •@hypolite I don't think you understood _anything_ of what you replied to.
Does that happen to you often?
Hypolite Petovan
Als Antwort auf Troed Sångberg • • •Troed Sångberg
Als Antwort auf Hypolite Petovan • • •Resuna
Als Antwort auf Resuna • • •Large language models aren't even based on the linguistic centers of the brain, they're based on the optical cortex, because that was the part of the brain most amenable to study when the somehat dated model of the neuron they use was developed in the '50s.