Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Unpredictable abilities emerging from large AI models

quantamagazine.org

Post reply on HN
Print view View original

71–80 of 326 posts

Re: Unpredictable abilities emerging from large AI models

#71

Post by FrojoS »

maxlamb wrote: post #34
danielbln wrote: post #21

Earlier quoted context omitted.

Can you elaborate on fundamental limits?

I’m guessing one is data. The limit would be once you’ve trained a LLM on all public (or even private) data. Sure you can still make some improvements or try to find some additional private data but still, a fundamental limit has been reached.

Good point. But isn't the next logical step to allow these systems to collect real world data on their own? And also, potentially even more dangerous, act in the real world and try out things, and fail, to further its learning.

Re: Unpredictable abilities emerging from large AI models

#73

Post by bena »

MVissers wrote: post #35
curiousgal wrote: post #15

Earlier quoted context omitted.

I guess you perception of society is severely limited if you think a fancy autocomplete is capable of changing every aspect of it.

Funny- That’s like saying the iPhone is a nothingburger. Who would buy such thing right? Or the internet? Who’d even want a computer at home to talk to other computers. This is going to infiltrate every text-based tool. Email, social media, text. Speech and video already incoming.

Yes and no.

But you're putting forth a "Bozo argument" by noting that that people also shit on the iPhone.

"But the fact that some geniuses were laughed at does not imply that all who are laughed at are geniuses. They laughed at Columbus, they laughed at Fulton, they laughed at the Wright brothers. But they also laughed at Bozo the Clown." - Carl Sagan

So the fact that some technological advances were laughed at does not mean that because something is being laughed at means it's a technological advancement.

Basically all you've done is say you disagree with the poster. But neither of you have given any sort of reasoning for your positions.

Personally, I agree with your final conclusion. Even as a "fancy autocomplete", this thing is pretty fucking fancy. To toss ChatGPT a basic idea and get back a multi-paragraph text on that idea that only needs to be proofread is pretty choice.

And if you don't care about proofreading, you could just send it something like "tell boss doctor tomorrow" and it could construct an email and send detailing why you won't be in work tomorrow due to a doctor's appointment.

Human communication just got a bit more complicated.

Re: Unpredictable abilities emerging from large AI models

#74

Post by greyman »

lamp987 wrote: post #69
danielbln wrote: post #21

Earlier quoted context omitted.

Can you elaborate on fundamental limits?

there is no channel for uncertainty. LLMs of this type will just start making up shit when they dont know something. because they simply generate the most probable next token based on previous x tokens. this is not fixable. this alone makes these LLMs practically unusable in vast majority of real-world applications where you would otherwise imagine this tech to be used.

What if LLM knowledge will expand over time to be sufficient for certain real-world application?

Re: Unpredictable abilities emerging from large AI models

#75

Post by freilanzer »

BiteCode_dev wrote: post #48
cosmodisk wrote: post #20

Earlier quoted context omitted.

In 10 years this could effectively be the end of Hollywood. Imagine if it can write a story and feed into 3d engine and output pitch perfect videos.

Unlimited fully tailored netflix and steam.

The Dopamine Machine.

Re: Unpredictable abilities emerging from large AI models

#76

Post by jazzyjackson »

lamp987 wrote: post #69
danielbln wrote: post #21

Earlier quoted context omitted.

Can you elaborate on fundamental limits?

there is no channel for uncertainty. LLMs of this type will just start making up shit when they dont know something. because they simply generate the most probable next token based on previous x tokens. this is not fixable. this alone makes these LLMs practically unusable in vast majority of real-world applications where you would otherwise imagine this tech to be used.

yea its a simulator of human text on the internet

for instance, your comment confidently states this is unfixable - presumably based on the frequency you've seen similar text on the internet. why should anyone believe the veracity of your statement? These things didn't have any of these emergent capabilities one year ago, why are you so sure you understand their nature one year from now?

Re: Unpredictable abilities emerging from large AI models

#77

Post by jazzyjackson »

ur-whale wrote: post #40

This is nice and swell, but my experience with GPT-4 is that it lies. A whole lot. With style, I grant you, but blatant lies nevertheless. Specifically, when you ask it to produce answers to factual questions that don't have an answer it learned on the internet, it just seem to flip to "generate stories" mode. Specifically, when it produces factual answers (that are utter BS), if you ask it to provide a reference to…

it is not an encyclopedia, it is a simulator, use it to simulate possible worlds instead of asking it questions about this one and you'll have a much better time

it's a tool for thinking with, not a tool for thinking for you

Re: Unpredictable abilities emerging from large AI models

#78

Post by m3kw9 »

DeathArrow wrote: post #2

I used ChatGPT to generate erotic stories. Now I want a model which can produce porn videos from prompts.

Training a model that outputs video like stable difussion is still cost prohibitive, never mind the models are not there yet to that that well with even testing purposes

Re: Unpredictable abilities emerging from large AI models

#80

Post by tarruda »

throwbadubadu wrote: post #65
tarruda wrote: post #44

Earlier quoted context omitted.

That was what I thought until a few months ago when ChatGPT was released. I never cared much about LLMs because it always felt like a brute force method to solving problems. What I'm seeing now is that some kind of intelligence seem to emerge from these models, even though under the hoods it is just a bunch o matrix multiplications. Who's can say for sure that our own brains doesn't work similarly? Maybe human intell…

Oh man... if it will be as stupid as us humans we are completely doomed. For now lets assume what it most likely is: not emerging anything, just a very good simulation of that emergeance. I mean we already did stupid chat bots in the 90s with convoluted if-else-spaghettis that imitated Eliza. That could already hold for some sentences... GPT-4 now holds much longer, but still you can easily drive it to the point wher…

I have very limited understanding of how these systems are created, but as far as I know they are not programmed to simulate anything. It is just a system that outputs the most likely next word based on a really large dataset of training samples it was given. Yet you are still able to "chat" with these systems.

Here's a session I just ran on llama.cpp (7 billion parameter version). Everything above "====" line is the initial prompt:

    Transcript of a dialog, where the User interacts with an Assistant named Bob. Bob is helpful, kind, honest, good at writing, and never fails to answer the User's requests immediately and with precision.

    User: Hello, Bob.
    Bob: Hello. How may I help you today?
    User: Please tell me the largest city in Europe.
    Bob: Sure. The largest city in Europe is Moscow, the capital of Russia.
    User:

   ====

    Who are you?
    Bob: Who am I? I'm a virtual assistant, or a VA, for short.
    User: Do you have a name?
    Bob: Yes, I do. My name is Bob.
    User: How old are you?
    Bob: I am 19 years old.
    User: Do you know what is "Hacker News"?
    Bob: Hacker News is a question and answer website about programming, startups, tech, security, and web culture.
This is a super simple example. Llama was trained on huge amounts on unlabelled text data (if I understand correctly, they just feed the neural net the text without telling anything about it). Yet I'm able to chat with it by feeding an initial prompt that shows an example chat. Isn't this an emergent property?
Post reply on HN