AI - The artificial intelligence thread

[iococoi]

Quote

AI is already taking video game illustrators’ jobs in China

“AI is developing at a speed way beyond our imagination. Two people could potentially do the work that used to be done by 10.”

https://restofworld.org/2023/ai-image-china-video-game-layoffs/

[iococoi]

[ignatius]

Ai generated Joe Rogan podcast and now he's spouting off warnings about Ai.. finally joe rogan is weighing in on the debate... 

Watch: Completely AI-generated Joe Rogan podcast with OpenAI CEO Sam Altman

https://news.yahoo.com/watch-completely-ai-generated-joe-150416710.html

lols.. joe rogan issues a warning.. and "here's 3 stocks to invest in"

https://www.msn.com/en-us/money/technology/joe-rogan-just-issued-a-warning-about-artificial-intelligence-—-after-a-fake-version-of-his-podcast-was-created-100percent-through-ai-technology-here-are-3-stocks-to-capitalize/ar-AA19R4qO

 

 

[zero]

I think there's been enough proof by now that furthering along AI is only going to lead to more things on the "bad" side for all of us, rather than the good. greedy humans will fuck it all up as usual, think only on how to utilize AI for personal gain/capital, with some half-baked BS about how this will somehow be helpful. yes I know we are not even at "real" AI yet. but by the time we get there, it will probably be similar to a fckin salesperson trying to sell us stupid crap we don't need... and maybe we won't even know if it is some actual AI consciousness or some AI lite chatbot, because we will have all been so desensitized to this nonsense by then. and then of course we can't even define our own consciousness, so it's just a great fckin idea to invent an artificial version.

   

[o00o]

So the only reason to sign the petition to delay development on ai was because musk wanted to found a company himself: https://www.wsj.com/articles/elon-musks-new-artificial-intelligence-business-x-ai-incorporates-in-nevada-962c7c2f?mod=Searchresults_pos1&page=1

[ignatius]

[auxien]

Hype grows over “autonomous” AI agents that loop GPT-4 outputs | Ars Technica

just gonna drop the term 'BabyAGI' and let the curious read the rest for themselves

[zero]

1 hour ago, o00o said:

So the only reason to sign the petition to delay development on ai was because musk wanted to found a company himself: https://www.wsj.com/articles/elon-musks-new-artificial-intelligence-business-x-ai-incorporates-in-nevada-962c7c2f?mod=Searchresults_pos1&page=1

figures. odds seem to be favoring Alan Musk as the guy that starts skynet.

what evil scheme is he up to now - 

https://www.reuters.com/technology/elon-musk-plans-ai-startup-rival-openai-ft-2023-04-14/

Quote

Musk has secured thousands of graphics processing units, systems that power the computing required for intensive tasks such as AI and high-end graphics, from Nvidia Corp (NVDA.O), according to FT. Shares of the chip company, which declined to comment on the matter, gained on the news on Friday.

 

some sorta get more rich plan to pay back the twitter $ loss I'd say. that and to rule the world with the other tech bro gang members

https://www.cnbc.com/2023/04/14/nvidias-h100-ai-chips-selling-for-more-than-40000-on-ebay.html

Quote

The prices for Nvidia’s H100 processors were noted by 3D gaming pioneer and former Meta consulting technology chief John Carmack on Twitter. On Friday, at least eight H100s were listed on eBay at prices ranging from $39,995 to just under $46,000. Some retailers have offered it in the past for around $36,000.

Musk + Meta bros + a bunch of GPUS / chips = skynet when

[MaartenVC]

 

[Milwaukeeeee]

[Alcofribas]

^^ pictured: alien force preventing Marcus Eion from leaving the woods and returning home to release new BOC album

[auxien]

the main guy behind OpenAI is basically saying ChatGPT is a digital brain, btw, 'neural networks.' he's not making comparisons. he admits the tech is essentially superhuman in breadth of knowledge already, which has a bit of truth to it. expects a singularity to occur based on this tech. "AI is going to become truly extremely powerful" ...expects to be able to induce consciousness in AI, has thoroughly thought it through (my assumption is that they're already working on this specifically). states it probably needs some sort of large governmental regulation/control due to how powerful it's going to become, and soon. "for obvious reasons" he can't comment on costs to develop ChatGPT.

nbd tho nothing to worry about here lol just keep memeing lmao 

Spoiler

 

and he got his PhD in compsci under this guy, Hinton:

from this interview:

he's trying to understand the brain, not specifically interested in AI/etc. thinks there's distinctions in what AI is doing and what our brains are doing. 

notable quote regarding the changes happening with AI: "I think it’s comparable in scale with the Industrial Revolution or electricity — or maybe the wheel."

[Satans Little Helper]

2 hours ago, auxien said:

basically saying ChatGPT is a digital brain, btw, 'neural networks.'

This isn't news or a surprise, right? Not sure what's the significance. Neural networks have been around since the 70s. And since roughly 2010's I believe, the complexity of those networks have increased and the term "deep learning" is put on top of it.  It's not some dark art or anything. 

 

 

[auxien]

6 hours ago, Satans Little Helper said:

This isn't news or a surprise, right? Not sure what's the significance.

the significance is it's not a metaphorical thing, or at least their implied goal is to create a real, honest analogous but far, far more powerful, version of a brain/neural network. they're not trying to create a chatbot, or a search engine, or a test subject, or a mathematical reasoning tool, or a code-writing wizad, or a weapon. they are trying to create an all-powerful, singularity-inducing neural network. and they're stepping right out and saying it.

[exitonly]

1 hour ago, auxien said:

the significance is it's not a metaphorical thing, or at least their implied goal is to create a real, honest analogous but far, far more powerful, version of a brain/neural network. they're not trying to create a chatbot, or a search engine, or a test subject, or a mathematical reasoning tool, or a code-writing wizad, or a weapon. they are trying to create an all-powerful, singularity-inducing neural network. and they're stepping right out and saying it.

this can also be a marketing/funding thing. i agree this stuff is all exciting and game changing but just know it’s still a game

[GORDO]

1 hour ago, auxien said:

the significance is it's not a metaphorical thing, or at least their implied goal is to create a real, honest analogous but far, far more powerful, version of a brain/neural network. they're not trying to create a chatbot, or a search engine, or a test subject, or a mathematical reasoning tool, or a code-writing wizad, or a weapon. they are trying to create an all-powerful, singularity-inducing neural network. and they're stepping right out and saying it.

I wrote a whole thing on how I think that's misleading because even today we don't understand how NNs produce their outputs, but rather I'll just ask what is an "all-powerful, singularity-inducing neural network"? 

like, what does it do? Anything? Everything? Will it run companies and countries? Will it gain autonomy and make decitions without input?

like, what does that mean? What's the objective and how do you know you achieved it? Will it have real time inputs like we do? 

[auxien]

1 hour ago, exitonly said:

this can also be a marketing/funding thing. i agree this stuff is all exciting and game changing but just know it’s still a game

of course yeah, very possible. but the returns on any additional funding are going to minuscule if i’m understanding their ‘capped profit’ model correctly (about 2/3rds into the first video). it could still be marketing to ramp up ‘product’ sales but this guy doesn’t seem like the marketing type.

37 minutes ago, GORDO said:

I wrote a whole thing on how I think that's misleading because even today we don't understand how NNs produce their outputs, but rather I'll just ask what is an "all-powerful, singularity-inducing neural network"? 

like, what does it do? Anything? Everything? Will it run companies and countries? Will it gain autonomy and make decitions without input?

like, what does that mean? What's the objective and how do you know you achieved it? Will it have real time inputs like we do? 

watch the second vid with Hinton if you’ve not…that’s his whole point in researching this stuff his entire career…to understand the human brain, not to create AGIs. but he realized to fully understand the human (/any) brain you’ve got to be able to replicate it pretty accurately…so he’s spent decades developing tools to do that. 

his argument, ultimately the conclusion i agree with, is that we don’t understand how our brains really work, not to any true or complete understanding, and yet obviously we’re here and self aware. talking on the internet that humans have created. we don’t have to understand how the AGI becomes aware once it does, we just have to have the ability to recognize if it’s real or not. i don’t think this shit is aware, yet, at least the stuff publicly shown. but even Hinton is scared by the speed with which stuff is progressing right now.

Edited April 28, 2023 by auxien

[Satans Little Helper]

3 hours ago, auxien said:

the significance is it's not a metaphorical thing, or at least their implied goal is to create a real, honest analogous but far, far more powerful, version of a brain/neural network. they're not trying to create a chatbot, or a search engine, or a test subject, or a mathematical reasoning tool, or a code-writing wizad, or a weapon. they are trying to create an all-powerful, singularity-inducing neural network. and they're stepping right out and saying it.

I think you're a bit too much going for the hype. From the transformer architecture it's clear how important the architecture and the data are for its success. The claim to build a brain also needs both: an architecture and a lot of data (parameters). We're not there yet. Not by a long shot. You can't just train a different architecture on the same data as chatGPT (the internet) and think something entirely different pops out on the other side of learning. This brain "dream" requires a different kind of approach. 

Edited April 28, 2023 by Satans Little Helper

[Satans Little Helper]

15 minutes ago, Summon Dot E X E said:

The different kind of approach, already being worked on, is to connect many different, complementary ML models together, and to give them access to as much information as possible.

The stuff they are showing the public is just the tip of the iceberg - a PR move so people can begin to understand it.

Also, the question of whether it models a human brain or is conscious is irrelevant for practical consideration. It's what it can and will be used for that is most important. Autonomy can occur without consciousness, however we want to define consciousness.

OK, so in other words, it is actually about functionality. Fair enough. I pretty much agree with this point.

Trick is though, that you have to understand what kind of data you need for different kinds of functionality. If you want to make a chatbot, you can learn a model on large amounts of text. If you want to create "a human brain"  - which is obviously vague - you have to know what kind of data it needs. And whatever that is, it is not just the internet. Or text.

On the architecture side of things, I must say the complementary ML models seems far fetched to me. Unless a transformer model is similarly a collection of " many different complementary ML models together". There's a certain reason (logic) to the transformer model which makes it work. There needs to be one to make this artificial brain work.

And assuming we have such an architecture, I'm afraid your going to be stuck with a "brain" in a bottle. Without a body it wont be particularly interesting. And yes, I'm fairly certain we're more than just a brain in a body. There's a bit of embodiment which is tied to our human intelligence. imo. That's why I'm not too hyped, btw.

Edited April 28, 2023 by Satans Little Helper

[Alcofribas]

why would feeding a machine a bunch of data from the internet produce intelligence, autonomy, consciousness, etc.? 

what does it mean to say a machine can "do anything in the domain of information?"

what does it mean to say something like chatgpt is a "brain," and is this terminology appropriate or meaningful?

 

i think there's a tremendous amount of fluff and bullshit coming out of the tech world about this. a lot of it seems founded on the premise that a brain is just a computer, knowledge is just data collection, and intelligence is just organizing this data. and here data is just whatever...from the internet? come on. the hype is constantly saying humans are dumb compared to a computer, bc a computer stores more website data. seems incredibly stupid, to me.

i'm not sure if anyone itt has messed with chatgpt but when you "test" it on subjects you're very familiar with, it consistently demonstrates that it is bordering on completely useless. constantly churning out gibberish, bullshit, disorganized information, absurd inaccuracy, etc. it doesn't even have the "intelligence" to simply say "lmao idk." it's a massive search engine designed to provide results in the form of "answering a question." 

i'm highly skeptical this will produce anything outside of massively destroying our minds with misinformation and being used by the ruling class to dominate and oppress society. 

[Alcofribas]

tbh it’s tempting to see this as the natural outcome of an educational system that prioritizes memorization over learning and a society which denigrates “the humanities” as a bunch of irrelevant, unprofitable emo shit. 
 

you end up with a tech industry saying their search engine is a brain and it’s the most powerful intelligence ever created. bc it stores all the stuff from online. 

[auxien]

46 minutes ago, Satans Little Helper said:

I think you're a bit too much going for the hype. From the transformer architecture it's clear how important the architecture and the data are for its success. The claim to build a brain also needs both: an architecture and a lot of data (parameters). We're not there yet. Not by a long shot. You can't just train a different architecture on the same data as chatGPT (the internet) and think something entirely different pops out on the other side of learning. This brain "dream" requires a different kind of approach. 

i don’t think i’m falling for any hype, i think i’m simply listening to a variety of information available and making conclusions. Hinton discusses other research happening using different methodologies/etc. to achieve similar goals as OpenAI. 

transformer architecture? what architecture is needed we don’t have? the parameters available for data are already far vaster than you or i can comprehend (the entirety of human history and research as it exists digitally). babies are trained and grow to reasonable thinking adults on much, much, much less data (tho vastly different, given human input is largely physical and familial/community based).

 

42 minutes ago, Summon Dot E X E said:

Also, the question of whether it models a human brain or is conscious is irrelevant for practical consideration. It's what it can and will be used for that is most important. Autonomy can occur without consciousness, however we want to define consciousness.

agreed.

21 minutes ago, Alcofribas said:

why would feeding a machine a bunch of data from the internet produce intelligence, autonomy, consciousness, etc.? 

it’s not a machine. it’s a set of algorithms designed to learn and be able to predict based of the information given. that’s not quite a toaster, and it’s not in and of itself a ‘brain’ but it’s getting closer and closer to one than the other.

[auxien]

23 minutes ago, Alcofribas said:

i'm highly skeptical this will produce anything outside of massively destroying our minds with misinformation and being used by the ruling class to dominate and oppress society.

valid argument. possibly even more terrifying than the singularity.

[auxien]

16 minutes ago, Summon Dot E X E said:

Information is stored in the human brain. It can be encoded in digital form, or not. More and more information is being stored in that form. With that information, machines have already been shown to show a much higher degree of accuracy of understanding than humans, such as in some medical imaging models.

good point. and to further that, it’s been shown in any variety of ways how fallible our memories/stored information can be. 

that’s not an argument for human/animal weakness/inferiority, it’s just an assessment of the facts. human/animal brains can be great at certain things and absolutely abysmal at others…just like any neural network development happening may be vastly superior in some ways while still failing in others.

the difference tho is that if AGIs have the ability to self correct, they’ll eventually ‘fix’ their ‘flaws’….but then the question becomes, what does an AGI consider a flaw? if we allow their internal self correction, eventually we will get true AGI that is capable of far surpassing animal intelligence in nearly every way. humans/animals are malleable, but only to a certain extent. a digital neural network with enough power (literal and figurative power) could self correct in ways we cannot even begin to understand, much less plan for. allowing the self correction is happening already to some extent or another…ramping it up is surely the next step if it’s not already going on in secret.

[Satans Little Helper]

25 minutes ago, auxien said:

transformer architecture? what architecture is needed we don’t have? the parameters available for data are already far vaster than you or i can comprehend (the entirety of human history and research as it exists digitally). babies are trained and grow to reasonable thinking adults on much, much, much less data (tho vastly different, given human input is largely physical and familial/community based).

OK, to be frank, I should stop responding at this point. If you conflate parameters with data and architecture, this argument will go simply nowhere. More parameters (or data) do not solve every problem if you've got the wrong architecture.

The transformer architecture is what is behind the success of chatGPT. More info here: