Corollary: anyone who thinks LLMs will give us AGI - regardless of academic or professional experience and expertise - either doesn’t understand how LLMs work, or is intentionally lying.
Comment on Major shifts at OpenAI spark skepticism about impending AGI timelines
cygnus@lemmy.ca 3 months ago
LLMs will not give us AGI. This is obvious to anyone who knows how they work.
gravitas_deficiency@sh.itjust.works 3 months ago
doodledup@lemmy.world 3 months ago
Maybe it can. If you find a way to conver everything to text by hooking in different models, the LLM might be able to reason about everything you throw at it. Who even defines how AGI should be implemented?
kia@lemmy.ca 3 months ago
The LLM is just trying to produce output text that resembles the patterns it saw in the training set. There’s no “reasoning” involved.
EnderMB@lemmy.world 3 months ago
A LLM is basically just an orchestration mechanism. Saying a LLM doesn’t do reasoning is like saying a step function can’t send an email. The step function can’t, but the lambda I’ve attached to it sure as shit can.
ChatGPT isn’t just a model sat somewhere. There are likely hundreds of services working behind the scenes to coerce the LLM into getting the right result. That might be entity resolution, expert mapping, perhaps even techniques that will “reason”.
Source: Working on this right now.
0laura@lemmy.world 3 months ago
they’re very very anti ai and crypto. I understand being against those, but lemmys stop caring about logic when it comes to those topics.
aodhsishaj@lemmy.world 3 months ago
You might be interested in Nim then when you get a chance. Talk about orchestration
doodledup@lemmy.world 3 months ago
You’re doing that too from day one you were born.
MentalEdge@sopuli.xyz 3 months ago
The “model” is static after training. It doesn’t continuously change in response to input, and even if it did, it would do so at a snails pace. Training essentially happens by random trial and error, slowly evolving the model towards a desired result. Human minds certainly do NOT work that way. Give a human a piece of information, and they can comprehend and internalize the relevant concepts in one go. And the actual brain is physically, permanently, altered through that process.
Once a model is trained, however, “memory” takes the form of tacking on everything the model has received and produced so far onto its input, each time it needs to output something more within that context. Each output hence become exponentially heavier to produce. The model itself no longer changes in any way beyond this point.
And, the models are all chronically sycophantic. If reason was involved, you’d not be able to just tell one to hold some given opinion. They’d have a developed idea of “reality” based on their dataset, and refuse to entertain concepts opposed to that internal model.
Once you get an LLM to hold a position, which you can do by simply telling it to, getting it to change should require a sane train of convincing logic. In reality, if you tell an LLM to defend a position, getting it to “change it’s mind” takes the form of a completely arbitrary back and forth that does not need to include any kind of sane argument. It will make good arguments, because it’s likely been trained on them, but your responses to it can be damn near complete gibberish, and it WILL eventually work.
Compare that to the way a human has to be convinced to change their mind.
Reasoning out concepts to come to conclusions isn’t something LLMs actually do, because again, the underlying model is static. All that’s actually happening is that the contents of the context are being altered until the UNCHANGED model produces an opposite response when fed the entire conversation so far as an input. Something which occurs every time it needs to produce new output.
LLMs can “reason” only in the sense that if you give one a thinking problem, it might solve it as long as the answer already exists somewhere in the data it was trained on. But as soon as you try to give it data to work with through your input, it can’t adapt. The model itself can’t evolve in response to what you are telling it. It’s static. It can only work with concepts that it has modelled during training, and even then it will make mistakes.
LLMs can mimic the performing of some pretty complex thinking problems, but a lot of the abilities required for something to become an AGI aren’t among them. Core among these is the ability for the model to alter itself based on input, and do so in a deliberate manner, getting it right within one or two tries.
In reality, training is brute-force process, not an accurate process of comprehension that nails down an understanding of a concept in one go.
nucleative@lemmy.world 3 months ago
This poster asked some questions in good faith, I don’t understand the downvotes when there’s a legitimate contribution to the conversation because that stifles other contributions.
Petter1@lemm.ee 3 months ago
And how does reasoning work exactly in the human body? Isn’t it LLM/LAM working together with hormones? How do you know that humans aren’t just doing something similar? Your mind tricks you about a lot of things you experience, how can you be sure, your "reasoning” is just sorta LLM in disguise?
MentalEdge@sopuli.xyz 3 months ago
The “how do you know humans don’t work the way machine learning does” is the wrong side of the argument. You should be explaining why you think LLMs work like humans.
Even as LLMs solve thinking problems, there is little evidence they do so the same way humans do, as they can’t seem to solve issues that aren’t included in their training data
Humans absolutely can and do solve new and novel problems without prior experience of the logic involved. LLMs can’t seem to pull that off.
MonkderVierte@lemmy.ml 3 months ago
Not at all. Btw, Autists reason entirely without words/language. Neurotypicals are capable of that too, but it’s more convenient for them to bridge over words in conscious reasoning.
LANIK2000@lemmy.world 3 months ago
Language models are literally incapable of reasoning beyond what is present in the dataset or the prompt. Try giving it a known riddle and change it so it becomes trivial, for example “With a boat, how can a man and a goat get across the river?”, despite it being a one step solution, it’ll still try to shove in the original answer and often enough not even solve it. Best part, if you then ask it to explain it’s reasoning (not tell it what it did wrong, that’s new information you provide, ask it why it did what it did), it’ll completely shit it self. There’s no evidence at all they have any cognitive capacity.
I even managed to break it once through normal conversation, something happened in my life that was unique enough for the dataset and thus was incomprehensible to the AI. It just wasn’t able to follow the events, no matter how many times I explained.
mke@lemmy.world 3 months ago
Except LLMs don’t actually have real reasoning capacity. Hooking in different models that can translate more of the world to text could give the LLM a broader domain, but not an entirely new ability beyond its architecture. That might make it more convincing, but it would still fail in the same ways as it currently does.
doodledup@lemmy.world 3 months ago
You’re doing reasoning based on chemical reactions. Who says it can’t do reasoning based on text?
mke@lemmy.world 3 months ago
If you genuinely think LLMs are anyway capable of even basic reasoning despite all ample evidence towards the contrary, I honestly don’t care about convincing you anymore. You’re asking for a miracle out of me—to explain consciousness itself, even—while you can just say “but there’s a chance” even though LLMs can’t get basic facts right.
MentalEdge@sopuli.xyz 3 months ago
Is language conscious? Is it possible to “encode” human thinking into the media we produce?
Humans certainly “decode” ideas, knowledge, trains of logic and more from media, but does that mean the media contains the components of consciousness?
Is it possible to produce a machine that “decodes” not the content of media, but the process through which it was produced? Does tmedia contain the latter in the first place?
How can you tell the difference if it does?
The more I learn about how modern machine learning actually works, the more certain I become that even if having a machine “decode” human media is the path to AGI, LLMs ain’t it.
conciselyverbose@sh.itjust.works 3 months ago
LLMs can’t reason about anything, ever.
anarchrist@lemmy.dbzer0.com 3 months ago
LLMs do not reason, they probabilistically determine the next word based on the words you prompt it with. The most perfect implementation of “AI” was the T9 predictive text system for dumb phones cmv.
MentalEdge@sopuli.xyz 3 months ago
And to have conversation, behind the scenes, each prompt gets the entire conversation so far tacked on.
The model itself is static, it doesn’t work like a brain that changes in response to stimulus, or form memories.
To converse about something, the entirety of an exchange is fed back into the model all over again each time it needs to produce a response. In fact, this cab happen several times over for each word in a response.
It’s basically an attempt at duct-taping the ability to form memories onto an otherwise static system. It works, but I don’t see how that way of doing it could ever land LLMs in the land of real consciousness.
It basically means these models “think” in frames, but each frame gets exponentially heavier to process, as it has to ingest every frame that came before.
mozz@mbin.grits.dev 3 months ago
OpenAI at least is now attempting to bolt on a “memory” by having the LLM spit out short snippets of what it might need to know later, which it then has access to when completing later prompts. Like everything else post-GPT-4, it seems fine but doesn’t work really all that well at what it is intended to do.
doodledup@lemmy.world 3 months ago
And you’re just a fancy electro-chemical reaction.
Who says that an LLM with complete access to the sensory world could not pass the Turing Test?
MonkderVierte@lemmy.ml 3 months ago
It’s already fact that the Turing Test only determines how much it can simulate human behavior. Nothing with intelligence to do.
capital@lemmy.world 3 months ago
I haven’t assumed that those who believe AGI will, at some point, come to be necessarily think that LLMs is exactly the tech which will get us there. Just that AGI is likely to happen because they don’t think there’s anything super special about the meat in our heads that makes intelligence possible and it should be able to be reproduced in other mediums.
kureta@lemmy.ml 3 months ago
I have been trying to convince my friend for weeks now. Not 24/7 of course. I try to explain how it works and we need at least another conceptually new method, this will never cut it. He says, “nobody could have predicted any of this, so you cannot be sure. You see, ChatGPT will take all the jobs in a couple of years.”
Rhaedas@fedia.io 3 months ago
LLMs alone won't. Experts in the field seem to have different opinions on if they will help get us there. What is concerning to me is that the issues and dangers of AGI also exist with advanced LLM models, and that research is being shelved because it gets in the way of profit. Maybe we'll never be able to get to AGI, but we sure better hope if we do we get it right the first time. How's that been going with the more primitive LLMs?
Do we even know what the "right" AGI would be? We're treading in dangerous waters.