ah, sorry for the confusion
1 Like
that makes a lot more sense, I should read more into it next time
1 Like
No, its literally just a large program hivemind that skims the web and stores data inside the data centers, and will answer questions, requests, etc. based off of what it is programmed to do.
The Image generation is part of itās programming, but it uses other works to train itself to be able to make images that dont look like cave paintings.
*also why it has its own āAI styleā, that makes it so easy to tell what is AI or not.
In short, its basically the equivalent to a soulless brain matrix that the big companies decided to make as a way to earn easy money on seamlessly ānewā tech, but actually AI has been around for a lot longer than 2 years.
Not sure this is the correct topic to ask that question though.
1 Like
I felt like an AI image generation programing of this caliber Adobe would have known and invested in it to be the first to own it outright, but apparently not.
Anyway thanks for the answer 
. Back to the topic.
The tech behind AI has been around for ages. But how it works simplified to the extreme is you feed a blank AI a ton of different either images, text conversation, information on certain topics, videos, etc. the AI learns what those things are from what itās given. You tell it make an image of a bunny and it only knows what a bunny should look like from images it has been fed. I think got 2.0 or so popularised the entire AI stuff and since then itās just hyped up and hyped up more
1 Like
Says people shouldnāt hate on LLMs, and then gives 4 really good reasons why we should. 
3 Likes
And the fact its going to cause a massive amount of job losses in the white collar sector.
Thats why the rich are pushing so hard for it. Less workers = more profit.
Not to mention they dont have to worry about unions rising up against them anymore. The social contract is in effect broken now.
3 Likes
It was an existing technology. In fact, it was one I was looking at using. I was learning how to make LLMs with PyTorch when ChatGPT came out. I had plans to integrate an LLM with a tool I was making, and then collect data from users over time to build a model.
Chat GPT was an interface for the GPT3 model, which was trained on the entire Internet. It was huge. OpenAI at the time was also open source. Which meant anyone could access what they did.
Everyone who had already been building LLMs no longer needed to spend time data training their models, because GPT3 had already done it at such a massive scale. It was like a sourdough starter. People could feed it new data and make their own models. They could take apart the source code and create new models from scratch.
No, but it became an arms race. Everyone was now trying to make the best LLM before everyone else to put everyone else out of business and win the economy. It was the speed at which they could train new data models, and the leveraging of an old technology - Video Cards - for a new purpose - creating LLMs. So suddenly there was a hardware arms race on as well as a software one. And those both require massive power and water.
An LLM just creates new things by combining things it has consumed.
It is not really a hivemind. Thatās giving it too much agency. It does not think. Itās a very fancy version of a human playing MadLibs, or the RNG in a video game - just amped up with a statistical model. Itās just generating statistically likely answers to questions based on the data it is fed.
To be clear the LLMs that create art are separate from LLMs that answer questions. They are trained on different data sets. When you ask ChatGPT to create an image for you, itās passing the request to another model trained to do that and returning the result. Though perhaps over time they will be combined.
Again, calling it a ābrain matrixā makes it sound like LLMs can think, reason, or make decisions. They cannot. They can make it pretty well, but they functionally are really good guessing machines. This article about an AI Dating Restaurant Experience that came out yesterday talks about how talking to an LLM bot is kinda like talking with a horoscope. It feels like itās talking to you based on the data it has, but the things it says could apply to lots of people.
We as humans anthropomorphize things, and so automatically do that with LLMs. And the companies pushing that encourage that by talking about āhallucinationsā as if LLMs think instead of just malfunction by generating a false statistical model. Because they do not know how to say I donāt know. They are statistical models. When you ask a question, they give the best answer possible. But they are not intelligent and they are creating a false answer because itās statistically probable to them - which they cannot distinguish from truth.
Have you ever talked to a drunk person who is convinced they are not drunk, and trying to convince you that they are not drunk? They believe it is true. LLMs are the same way. Except instead of arguing with you, an LLM just takes in the data that you say they are wrong and adds it to the statistical model to give you a new answer that also is just statistically probable.
Companies also help us anthropomorphize LLMs by talking about AGI (Artificial General Intelligence) and how LLMs will take over the world and destroy it like the movie Terminator. Except that LLMs cannot make decisions. Which is why āAgentic AIā is so scary. It is basically giving a bunch of gambling machines the ability to affect the outside world based on predictions of the statistically probable way something should be done.
5 Likes
some models will act autonomously, but it is based off of training and yes outcome prediction.
āThank you for the correction, much appreciated, my response wasnt very well structured,
shoot this sounds like llm text. . . .
1 Like
Rejoice LLM haters! A considerable chunk (if not all) of Claude Code source code leaked out to public due to a rookie security mistake by none other than Anthropic themselves. And it happened on April 1st, well a day before. Somebody immediately vibe coded and published a Python port of the whole thing, against which no copyright infringement actions can be taken. To cite the famous ai-bro adage: āWell bruh, if itās on the internet - it a fair game for takingā. The irony factor with all thisā¦ The comedy practically writes itself here
6 Likes
Yeah, I do agree with this. It is a little bit like the gun debate.
āGuns arenāt the problemā.
āBut if you give a person a gun they are more likely to shoot someone.ā
āBut guns are often used for protection of families or propertiesā
āOnly because the attacker also has a gunāā¦
And it goes on forever.
Personally I think the UK is better off without the widescale population being armed. However if it did allow guns, I would definitely get one. So perhaps the world would be better without the widescale use of LLMs. And perhaps the only way to stop them would be to enforce the laws governing copyright, perhaps even extending copyright to say it includes usage of said works in training materials, public or private.
I donāt know, I donāt think it is clear cut. And I donāt think everything LLM related is bad. At some point though returns are going to be required for the investments, and I still donāt see where they are coming from.
It will be much worse when robotics reach a point of replacing humans. I know we get closer to this every day, where three people farm thousand acre farms, where two warehouse staff manage 100,000ft warehouses etc etc. Wait though until there are no delivery driver jobs, no parking enforcement officers, no police, no armies, no receptionists or doorman, no factory workers, no cleaners, no binmen, all replaced with humanoid robots. What then?
Sadly, I agree with you. And even more sadly, I think people have forgotten what war is really like, and that there are no benevolent dictators.
2 Likes
I was just reading about this today, but that video is extra hilarious. Definitely a really good explanation for those who want to understand a bit more about LLMs too.
Itās a lot like the gun debate. I originally drew the comparison, but then decided that might come across as too polarizing and deleted that part. Glad you saw where I was going though.
Sadly, I think you will find that armies are the last to go, because human lives are economically cheaper than robots on the battlefield.
3 Likes
Thank you! This is the only sensible comment in this thread.
TLDR;
- Anger directed at AI is nonsensical. The root of the problem isnāt the technology but how itās applied, in the same way that a gun in itself isnāt an issueāitās all about how it is used. The way LLMs are used today is a byproduct of the cultural, economic, and political preconditions of the presentāin other words, you should really be pointing a finger at capitalism and atomistic individualism.
- The AI bubble is not a bubble in any sense of it being a āpassing crazeā. Anyone who thinks itās a bubble in this sense may as well wave a flag that says āI donāt know what Iām talking about but Iād like to voice my feelings anyways.ā The underlying technology is sufficient to create sentient artificial life, if orchestrated in the right ways, and thatās going to change everything.
1 Like
Thatās nowhere near certain, and nothing is pointing in that direction, unless you have some reputable sources that a glorified word guesser is going to turn into sentienceā¦
Thinking that LLMs are anything like how sentient beings think is massively misunderstanding either how human brains work, or how LLMs work, or both
6 Likes
Youāre trolling, right?
8 Likes
Anger is a reaction to something in life feeling unfair. Itās not nonsensical. You cannot logic away emotions. We are not Vulcans. We can deal with those emotions appropriately, but to start by telling people not to have emotions isā¦well an odd way to get oneās point across in my opinion.
This was already addressed, so Iām not going to repeat myself: If you hate LLMs, there's hope - #6 by dragonforge-dev
Thatās a very nice philosophical point of view, and I agree with it, but pointing a finger at -isms does not move the needle.
No one said that. You are misunderstanding.
We are talking about an economic bubble. Very specifically likening it to the DotCom economic bubble in March 2000. Some of the context of this thread may be drawn from this one: Is the AI Bubble about to burst? as well as a few others on this forum.
In context of your next sentence, I find this amusing. First, because no one was saying that, and second because you clearly do not understand the underlying architecture of LLMs.
No, itās not. LLMs are spicy autocomplete machines. They are statistical models that make statistical guesses. They have no sentience. They do a good job of appearing sentient, until you run into an uncanny valley. AGI is not going to happen with LLMs. I discussed this at length in this reply: If you hate LLMs, there's hope - #32 by dragonforge-dev
I gotta agree with @normalized, you do appear to just be trolling.
9 Likes
At this point i dont know if he is XD
1 Like
I think the thing that really fascinates me about LLMs is that they proved that the Turing test wasnāt really relevant, turns out we attribute sentience to something that seems sentient, which is kind of scary
6 Likes
So for the bubble part, the current AI market is absolutely massively overinflated, partially as any novel market is going to be, before the long term players establish themselves, but also because of unsustainable investment practices and speculation, it will bust, the question is more in what way and for what reason, it could be because they technology just fizzles out because it stops innovating and people grow bored of it, or because the over speculation dies off and lots of money falls out of the system
4 Likes
If by āseen asā you mean āpeople mistake it for sentienceā then yes, but it isnāt sentient
5 Likes