AI output indistinguishable from drake and the weeknd music

This is for discussions about news, politics, sports, other games, culture, philosophy etc.
User avatar
Latvia harcha
Gendarme
Posts: 5136
Joined: Jul 2, 2015
ESO: hatamoto_samurai

Re: AI output indistinguishable from drake and the weeknd music

Post by harcha »

iNcog wrote:
25 Apr 2023, 01:52
Does this mean, however, that a layman such as myself would be able to write their own programs? Even developed ones? I have some ideas but no real skills for execution.
You could already do that before depending on what you mean by "programs". Writing little python scripts to automate small tasks has always been approachable.
POC wrote:Also I most likely know a whole lot more than you.
POC wrote:Also as an objective third party, and near 100% accuracy of giving correct information, I would say my opinions are more reliable than yours.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

iNcog wrote:
24 Apr 2023, 23:13
Goodspeed wrote:
23 Apr 2023, 06:42
Musk is irrelevant. Yeah, Google seems to be losing to MS. Basically OpenAI is far ahead of everyone and MS partnered with them which was smart.
They were feeling that the big tech boom was going bust, so they pushed this with big fanfare to maintain the illusion that the big tech boom is not over. They have this big next thing coming out, stay tuned and keep buying shares. It's all about share prices.
The big fanfare is because it's a big deal. I actually don't think there's enough fanfare. People still seem generally unaware of this and if they are aware they underestimate the implications.
I'd feel more enthused if it gave me code that didn't require me to debug it. Sure it's just small little python scripts, but still.

I think I'm using it wrong when it comes to using it for coding that said.
Are you using GPT-4 (if so how?). It's miles better at this than the current version.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

iNcog wrote:
25 Apr 2023, 01:52
Does this mean, however, that a layman such as myself would be able to write their own programs? Even developed ones? I have some ideas but no real skills for execution.
More complex programs still require at the very least a human who understands what it should look like to guide the model. For now.
France iNcog
Ninja
Posts: 13236
Joined: Mar 7, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by iNcog »

Goodspeed wrote:
25 Apr 2023, 07:07
iNcog wrote:
24 Apr 2023, 23:13
Show hidden quotes
I'd feel more enthused if it gave me code that didn't require me to debug it. Sure it's just small little python scripts, but still.

I think I'm using it wrong when it comes to using it for coding that said.
Are you using GPT-4 (if so how?). It's miles better at this than the current version.
I give it a fuzzy, easily misunderstood description of what I want then ask it to write code. I have no doubt that this isn't the right way to get it to work. Which beggars the question, what is the right way?
YouTube: https://www.youtube.com/incog_aoe
Garja wrote:
20 Mar 2020, 21:46
I just hope DE is not going to implement all of the EP changes. Right now it is a big clusterfuck.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

I mean how did you get access to it. It's kind of a pain from what I understand
No Flag RefluxSemantic
Gendarme
Posts: 5996
Joined: Jun 4, 2019

Re: AI output indistinguishable from drake and the weeknd music

Post by RefluxSemantic »

Out of curiosity I keep trying to use chatgpt for work. It is honestly completely useless and significantly worse than just googling.

We will be getting gpt4 at some point, but I personally am not yet convinced. I wouldnt pay significant money for it.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

What about for GPT-5? Or 6? Or 12? Or... oh wait it killed us
France iNcog
Ninja
Posts: 13236
Joined: Mar 7, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by iNcog »

Goodspeed wrote:
25 Apr 2023, 16:08
I mean how did you get access to it. It's kind of a pain from what I understand
Oh my bad, I don't have access to the GPT-4 version. I was discussing the March 2023 version that is free to use for anyone.
YouTube: https://www.youtube.com/incog_aoe
Garja wrote:
20 Mar 2020, 21:46
I just hope DE is not going to implement all of the EP changes. Right now it is a big clusterfuck.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

Someone theorized that the reason humans are able to "ruminate" the way that we do is that we have loops in our brain's neural network, as opposed to current artificial neural networks which are forward only. Currently they're a one way street going input -> neural net -> output but if we can enable it to form loops in the neural net somehow, the way we have them, it could mean that it starts ruminating like we do. It could essentially go from a knowledgeable entity with very good intuition to a ruminating, thinking brain that can actually validate its intuition by giving it further thought. It's also no stretch to think this could make it conscious.

Just an intuitive, easy to understand way that these models could leap forward in capabilities, if they can overcome the technical challenges.
No Flag RefluxSemantic
Gendarme
Posts: 5996
Joined: Jun 4, 2019

Re: AI output indistinguishable from drake and the weeknd music

Post by RefluxSemantic »

From what I know from the development of children, human intelligence is at least partially hardwired.

For example, the theory of mind is something that develops in children at a very consistent age. Same for a few other things, like abstract thinking which develops somewhere between 12 and 15 generally.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

@RefluxSemantic I've been wondering about the limits of intelligence, in particular in the natural sciences like physics.

Say we make an AGI with the capability to learn all of human knowledge and has the ability to reason and think much like we do (or maybe even more efficiently than we do), but has a million or even a billion times the processing power of our brains. Basically imagine a system that is pushing the limits of intelligence.

While humanity, to get to computers for example, has needed many small leaps in technology to eventually get here, what is your intuition about what such an intelligence could do in the physics domain? Would it be able to make big leap inventions that would normally require many small leaps, such as fusion energy, space ships going at light speed and potentially beyond, etc? Or do you think we still don't have the knowledge required to "deduct" your way to such inventions even if your ability in doing that is basically limitless?

There is a world where we invent AGI, it improves itself until it's as intelligent as I described, and we can start asking questions like "How do we cure cancer?", "How do we generate energy as efficiently as possible?" etc and it could answer those questions immediately, though it may have a hard time communicating its solutions to us.
No Flag RefluxSemantic
Gendarme
Posts: 5996
Joined: Jun 4, 2019

Re: AI output indistinguishable from drake and the weeknd music

Post by RefluxSemantic »

Even dumb unlimited computing power would solve quite some difficulties. For example, in my work (integrated photonics) we would ideally do big FDTD simulations. This is a type of simulation that is essentially an exact solution to your problem (so its not just a model, you are really solving the physics of the problem. Ideally we would do this simulation for the entire parameter space to find the optimum. These simulations are unfortunately very slow and dont scale well with device size, so we have to be smart and solve our problems in 'intelligent' ways. So we make internal models to understand the behaviour, use approximate models that are faster to simulate, try to extrapolate findings etc. But with unlimited computational power, none of this would be necessary. And thats before adressing the intelligent aspect of your suggestion. So from an engineering perspective, an AI like this would be extremely useful. Im not very knowledgable about other fields of engineering, but I could imagine that this would be very useful for aerodynamics (itd be like an F1 team on steroids) and for those protein calculations that might allow it to quickly find new solutions.

However, realistically speaking these simulations are not going to be any faster by simply developing AGI (they are limited by the raw computing power that we can achieve). So in that sense, I do not see the big difference between an AGI aided by simulation tools and a human aided by simulation tools.

Maybe AGI could advance theoretical physics. Im not a theoretical physicist so I dont really know their flow. But I suppose that with AGI you could have millions of artificial theoretical physicists crunching on the current problems, which should be quite a bit faster than what we have now. Plus, they can be aware of all knowledge and thus dont have to specialize, though I am not sure how big of a limitation this is. I suppose the learning would still be similar, but it would be faster and all agents could communicate their findings instantaneously.

So ultimately I think the biggest impact would be that we could have millions of artificial scientists working on every problem. Look at what happened during covid when scientists got funded heavily, and then try to imagine what would happen if every single field of science had even more funding.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Dolan »

Increasing computational power helps solve problems that need to perform astronomically large simulations to predict a specific outcome, like in the case of protein folding. Here it can work because the problem can be broken down into fundamental physico-chemical dynamics that are well understood. What is left is to cover all the possible ways in which some molecules will get arranged in a given space, following a certain energy landscape. And even in such a case, while current AIs have a high prediction rate, it's not 100%, it's at most 80%. Which is significantly short of calling computational protein folding a solved problem.
This can still work because, again, there's actually no reasoning involved here, there are only mathematically computable possible arrangements in a given system of coordinates. And when it doesn't work, it's humans that have to go back to the drawing board and think why it didn't work and look for ways to refine the model to be able to cover those cases that elude prediction. So the actual thinking about how the model should be formulated and rendered in mathematical form to be translated into a computing environment is done by humans. It's not the AI that is able to take stock of the situation and realise there's a large gap between its predictions and what is found experimentally.

What I'm implying is that simply increasing data processing does not necessarily produce better insight into a phenomenon. And it doesn't produce insight into a problem that wasn't first mathematised and translated into a computational language by humans. The AI itself does not on its own understand the world using language and is not able to "make sense" of the patterns it "recognises". No, the problem is formulated by humans who know what to ask from the software and who formulate the problem in computational terms. So that they actually get results that make sense for humans.

Here's a classic example of how scientific reasoning is not just data processing or algorithmic thinking. Anyone with some interest in science probably at some point has read Einstein's special theory of relativity and the thought experiment with the train and the platform. There are many cool variations of this thought experiment, personally I like the one with the flashes of light emitted from the center of the traincar:
[There's] one observer midway inside a speeding traincar and another observer standing on a platform as the train moves past.

A flash of light is given off at the center of the traincar just as the two observers pass each other. For the observer on board the train, the front and back of the traincar are at fixed distances from the light source and as such, according to this observer, the light will reach the front and back of the traincar at the same time.

For the observer standing on the platform, on the other hand, the rear of the traincar is moving (catching up) toward the point at which the flash was given off, and the front of the traincar is moving away from it. As the speed of light is finite and the same in all directions for all observers, the light headed for the back of the train will have less distance to cover than the light headed for the front. Thus, the flashes of light will strike the ends of the traincar at different times.

Image
This thought experiment was meant to illustrate that there is no such thing as absolute simultaneity of spatially separated events. The timing of events will be perceived (and measured) within each observer's frame of reference.
This kind of thought experiment exposed at the time when it was conceived a mismatch between preexisting abstractions (theories) about the world and what could be experimentally verified, if all the presuppositions were correct (speed of light being a constant and all the velocities being added). It effectively pointed out that theories needed to be adjusted for the other presuppositions to be consistent with each other. It's a collision between abstractions and reality for which a solution could only be found by coming up with imaginary experiments that could not be algorithmically deduced from any pre-existing model or theory (which were incomplete, thus wrong). A mind, or some minds, had to take the preexisting theory and mentally test its predictions in a physically lived, imagined space, in order to come up with conclusions that revealed something about the acuracy of the theoretical model.

This sort of real, physical-world intuition is something that would be missing from any AI that could not imagine a situation like: what if instead of testing this idea with pulses of light emitted from the center of the traincar, we tested it by having electrodes connected at each end of the traincar (one at point A in the back, another at point B in the front) and having them collide with other electrodes from a parallel traincar that comes from the opposite direction; the light pulses that result from the colliding electrodes would propagate at the same speed in space towards the observer in the center of the traincar, but since the traincar was moving towards the front, the light pulse from the back of the traincar (point B) would arrive faster than the pulse from the front (point A), because it'd be "catching up" with the observer faster.
visual illustration of this imagined experiment
No Flag RefluxSemantic
Gendarme
Posts: 5996
Joined: Jun 4, 2019

Re: AI output indistinguishable from drake and the weeknd music

Post by RefluxSemantic »

In the previous hypothetical, we assumed AGI, so an AI with the logical reasoning of our top minds. Which is very, very far away.

I was initially impressed by chatGPT but slowly I am reaching the conclusion that for most practical purposes, it is absolutely useless. The problem is still the bullshitting. Allow me to share the following excerpt to show just how unreliable any of its answers are:
Spoiler


Sorry for the technical topic, but I mostly test it at work. Its acceptable that it doesnt succeed at the task. However, it is unacceptable that it claims to succeed and claims to base that succes on a source while it is in fact pulling the information out of thin air. When you try to confront it, it will double down and claim that it in fact isnt bullshitting you.

In practice, this means that I need to double check all of Bing's claims. So I might as well just put in the work myself. As a tool to extract information from the internet, it is still absolutely useless because of that.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

RefluxSemantic wrote:
30 Apr 2023, 08:32
In the previous hypothetical, we assumed AGI, so an AI with the logical reasoning of our top minds. Which is very, very far away.
Define very, very far? And how are you so sure?

It could take a century, but it could also take 2 years.

What's impressive about GPT-4 is that it showed a very big jump in reasoning ability, theory of mind and just general understanding compared to GPT-3.5. It will be useful to many, not necessarily to you, but the story (to me) is the potential.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Dolan »

Image

rofl
I think "taking over the world" will take a while
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

It's actually better at math than I am, but basic arithmetic is hard for these models. It knows what to do, but it makes mistakes simply adding or multiplying stuff. I asked GPT-4 to multiply these numbers and to explain how it's doing it, and it shows the correct method (multiplying every digit of the first number with every digit from the second etc) but the numbers it comes up with are just wrong.

It makes sense because it's trained to predict the next token, so it will accurately predict the method by which a human may have solved the problem, but the actual answer it can only guess. It's a limitation in the current architecture. It gets surprisingly close though, considering it's just guessing.

Goes back to my earlier point that it's a system with excellent "intuition" (superhuman, in fact), but no ability to check its intuition or do anything other than intuit the answers. Kind of like a human without the ability to think, who would only respond with the first thing that comes to mind.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

viewtopic.php?f=315&t=23930

I asked it about the sequence in that OP. Impressively (to me) it is able to deduce the exponential pattern. It's impressive not because it's a hard pattern to discover but because it shows that it actually understands mathematics to a degree that I didn't think would be possible with this architecture. Also, it understands what I'm asking in the second screenshot, which I also think is impressive. It shows that "understanding" itself is an emergent property.

Again some of its basic arithmetic is wrong, and it's missing that the n-th number is not just f(n-1) * 1.1, rather f(n-1) * 1.1 + 1, but the fact that it gets as far as it does is huge to me.

Image

I asked it for a function that doesn't rely on knowing f(n-1) after it told me to just repeat f(n-1) * 1.1 if I want to know f(100):

Image
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Dolan »

Rise of the Newsbots: AI-Generated News Websites Proliferating Online
https://www.newsguardtech.com/special-reports/newsbots-ai-generated-news-websites-proliferating/

NewsGuard has identified 49 news and information sites that appear to be almost entirely written by artificial intelligence software. A new generation of content farms is on the way.
Artificial intelligence tools are now being used to populate so-called content farms, referring to low-quality websites around the world that churn out vast amounts of clickbait articles to optimize advertising revenue, NewsGuard found.

In April 2023, NewsGuard identified 49 websites spanning seven languages — Chinese, Czech, English, French, Portuguese, Tagalog, and Thai — that appear to be entirely or mostly generated by artificial intelligence language models designed to mimic human communication — here in the form of what appear to be typical news websites.

The websites, which often fail to disclose ownership or control, produce a high volume of content related to a variety of topics, including politics, health, entertainment, finance, and technology. Some publish hundreds of articles a day. Some of the content advances false narratives. Nearly all of the content features bland language and repetitive phrases, hallmarks of artificial intelligence.
The last sentence is the money shot
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Dolan »

It's possible that this proliferation of AI-generated content will actually make content creators stop sharing anything for free and websites putting up more paywalls and filters to make sure their content will not be indexed and data-mined by AIs.
There's an ongoing strike in Hollywood right now, on a related note, as writers are seeing their role undermined by the streaming business model (Netflix) which outsourced lots of jobs to countries with cheaper labour. So all these writers who create content for TV shows are seeing their role diminished and turned into a gig economy job with lower pay. AI content creation adds another dimension to this conflict, as corporations might start using more AI tools for ordinary write-ups and hire cheap outsourced freelancers from abroad to just correct the bland AI output and add a human touch to it. Then content creation for the big media might become very cheap, making their business model more competitive relative to streaming businesses.

This might lead to a more closed future internet as concerns about which content is AI generated and which is human will spread like wildfire, making everyone doubt most of the content they see online. Then finding trust-worthy sources of information will become a struggle and many will fall for AI-generated digital slop.
There are some cool scenarios that could come out of this as the internet might undergo a medieval phase, in the sense that it would become very localised, with people seeking their own closed local bubble, in which they know everybody and can personally check who generated which content.
This would effectively crash the big media business model, as almost nobody would trust anything with a very wide area of circulation, since they'd suspect corporations are trying to cut costs and use AIs to generate their content for cheap, with some extra freelancing human touch on top, just to avoid having their content look too generic.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Dolan »

Image

one day, this is gonna perform surgery on you, let that sink in
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Goodspeed »

That's like saying "one day, this 4 year old is going to perform surgery on you [as a 4 year old]" knowing damn well that by the time he's doing surgery he's not going to be a 4 year old, he'll be a trained 30 year old. It's obviously not this baby version of AI that's going to cut into anyone, it's the future version that's way smarter than you at everything. I imagine surgery is something robots would be especially good at, having perfectly steady "hands".
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: AI output indistinguishable from drake and the weeknd music

Post by Dolan »

ImageImage
over for instahoes

Who is online

Users browsing this forum: No registered users and 10 guests

Which top 10 players do you wish to see listed?

All-time

Active last two weeks

Active last month

Supremacy

Treaty

Official

ESOC Patch

Treaty Patch

1v1 Elo

2v2 Elo

3v3 Elo

Power Rating

Which streams do you wish to see listed?

Twitch

Age of Empires III

Age of Empires IV