AI: The Rise of the Machines... Or Just a Lot of Overhyped Chatbots?

Fascinating

Some good points but nobody is ever going to use these offline because they will only ever be as good as the day you downloaded it. You personally need to train it further after that and it will only be learning from your own input. That might be fine for one or two functions that you are already proficient in and can train and tailor it yourself. It won't work very well at all for anything else.

These things will honestly need to be connected to the internet to be worth using, and when that happens they become distributed systems. That's what is way more likely. Those datacenters going tits up because these LLMs are distributed and use all the computing power of home PCs across the globe, aka the torrent or blockchain of AI. And even though the hardware in those datacenters is still highly useful and specialized there isn't enough profit in running them because you can't charge the premium (like he illustrates).

The software companies creating and providing these LLMs will not all go out of business, we'll just crown a few kings and the majority of people will use those LLMs in this way.

Khane said:

Some good points but nobody is ever going to use these offline because they will only ever be as good as the day you downloaded it. You personally need to train it further after that and it will only be learning from your own input. That might be fine for one or two functions that you are already proficient in and can train and tailor it yourself. It won't work very well at all for anything else.

These things will honestly need to be connected to the internet to be worth using, and when that happens they become distributed systems. That's what is way more likely. Those datacenters going tits up because these LLMs are distributed and use all the computing power of home PCs across the globe, aka the torrent or blockchain of AI. And even though the hardware in those datacenters is still highly useful and specialized there isn't enough profit in running them because you can't charge the premium (like he illustrates).

The software companies creating and providing these LLMs will not all go out of business, we'll just crown a few kings and the majority of people will use those LLMs in this way.

Click to expand...

Well, only as good as when you downloaded? yes, but if that covers your use case, why is that a problem?
Is your operating system more useful than it was 10 years ago because of all of it's "improvements"? Your browser? Has coding changed so much in the last 10 years that someone who could code 10 years ago couldn't code today? Writing?
And just because it's local doesn't mean that it can't process new info, just that it doesn't get new training. You can have it google for things, which might not technically be "offline" but that's entirely different than using a model that's fully controlled by a corp. You can always download whatever and point it at the offline data if that's a sticking point for some reason.

But even then, at least for the moment new open source models are still being created. So at least as long as companies are willing to do that, you can still run new local models.

I can't tell if you're being serious or not. The entire purpose of LLM's and AI is that it can be trained and "learn". What you are describing is a downloaded knowledge base, not an LLM, and those have existed for a long, long time.

Khane said:

I can't tell if you're being serious or not. The entire purpose of LLM's and AI is that it can be trained and "learn". What you are describing is a downloaded knowledge base, not an LLM, and those have existed for a long, long time.

Click to expand...

So you're saying Qwen 3.5 is a knowledge base? or is it useless because it doesn't live on Anthropic's server?
edit: and why would it be more useful because it lives on someone else's server? Just because their server is better than yours?

Control said:

So you're saying Qwen 3.5 is a knowledge base? or is it useless because it doesn't live on Anthropic's server?
edit: and why would it be more useful because it lives on someone else's server? Just because their server is better than yours?

Click to expand...

I said exactly what I meant to say, in the post you quoted. Did you watch the video in question because that's what I was commenting on, what the musician in the video's understanding of AI and its future is. He thinks the majority of people are going to download these LLMs and then go offline with them. That is so unlikely it may as well not even be discussed and it also invalidates their main function and ability to be trained en masse to a point where their learning capability is actually meaningful.

Khane said:

I said exactly what I meant to say, in the post you quoted. Did you watch the video in question because that's what I was commenting on, what the musician in the video's understanding of AI and its future is. He thinks the majority of people are going to download these LLMs and then go offline with them. That is so unlikely it may as well not even be discussed and it also invalidates their main function and ability to be trained en masse to a point where their learning capability is actually meaningful.

Click to expand...

Ok, so I just watched it, and I don't think that changes anything I said. Of course most people aren't going to run their own models though since most people are retarded. People would rather stream something than have their own media file too. That doesn't have anything to do with whether or not local ai can be useful. And "invalidates their main function"?? Have you actually used any of these things?

No idea if the availability of local models will have much of an impact on the big ai companies though. Companies are happy to throw lots of money at all things cloud so they don't have to deal with it themselves. I'm not sure the pace of current ai dev can continue. I imagine it will be like everything else where once someone wins, they'll just cut costs, increase prices, and make it shittier. However, what open models are doing is allowing competing companies to offer similar offerings for tiny, tiny fractions of the investment. If anything other than their insane spending dooms the big companies, cheaper competitors running open models is it.

Control said:

Ok, so I just watched it, and I don't think that changes anything I said. Of course most people aren't going to run their own models though since most people are retarded. People would rather stream something than have their own media file too. That doesn't have anything to do with whether or not local ai can be useful. And "invalidates their main function"?? Have you actually used any of these things?

No idea if the availability of local models will have much of an impact on the big ai companies though. Companies are happy to throw lots of money at all things cloud so they don't have to deal with it themselves. I'm not sure the pace of current ai dev can continue. I imagine it will be like everything else where once someone wins, they'll just cut costs, increase prices, and make it shittier. However, what open models are doing is allowing competing companies to offer similar offerings for tiny, tiny fractions of the investment. If anything other than their insane spending dooms the big companies, cheaper competitors running open models is it.

Click to expand...

Nobody said anything about not running LLMs locally. I said they won't run them offline and yes running them OFFLINE invalidates their main function. I even specifically mentioned what will likely happen in the post you quoted:

These things will honestly need to be connected to the internet to be worth using, and when that happens they become distributed systems. That's what is way more likely. Those datacenters going tits up because these LLMs are distributed and use all the computing power of home PCs across the globe, aka the torrent or blockchain of AI

Click to expand...

Khane said:

Nobody said anything about not running LLMs locally. I said they won't run them offline and yes running them OFFLINE invalidates their main function. I even specifically mentioned what will likely happen in the post you quoted:

Click to expand...

So what do you think their main function is? I can unplug my internet connection and a local model can still code for me, write for me, answer questions for me, run automations for me, generate images and video for me. I guess it can't look up today's weather or answer emails, oh no!

This is why I asked if you were serious. I literally, in the first sentence, explained why using them offline invalidates their main function, which is to keep learning.

You think AI already knows everything it needs to know. This is not how humans work, we don't just "go offline" and stop learning. AI is "Artificial Intelligence". The I is pretty important.

But I guess I understand why you think this way. You asked if writing software and code was different now than it was 10 years, the answer is yes, quite a bit. The LLMs are designed to keep learning, to keep up with human ingenuity and to try to behave like humans do. You have the same misunderstanding of AI most people do. And its why so many people think AI is already capable of taking high skill jobs.

AI does not write code for you, it doesn't understand how to write code, not the way humans do. If you try to throw an AI agent into a custom codebase it fails miserably until it gets trained on that code base. It reads through its own data "memory" and outputs what it thinks is an appropriate answer to your coding question, based on information it has absorbed and learned from humans writing it down and giving it access. And it's often not correct. So we correct it, collectively and it "learns".

If you still don't understand why an offline LLM invalidates its main function, I'm not sure what else to say.

Khane said:

This is why I asked if you were serious. I literally, in the first sentence, explained why using them offline invalidates their main function, which is to keep learning.

You think AI already knows everything it needs to know. This is not how humans work, we don't just "go offline" and stop learning. AI is "Artificial Intelligence". The I is pretty important.

But I guess I understand why you think this way. You asked if writing software and code was different now than it was 10 years, the answer is yes, quite a bit. The LLMs are designed to keep learning, to keep up with human ingenuity and to try to behave like humans do. You have the same misunderstanding of AI most people do. And its why so many people think AI is already capable of taking high skill jobs.

AI does not write code for you, it doesn't understand how to write code, not the way humans do. If you try to throw an AI agent into a custom codebase it fails miserably until it gets trained on that code base. It reads through its own data "memory" and outputs what it thinks is an appropriate answer to your coding question, based on information it has absorbed and learned from humans writing it down. And it's often not correct. So we correct it, collectively and it "learns".

If you still don't understand why an offline LLM invalidates its main function, I'm not sure what else to say.

Click to expand...

Maybe that's true in some nebulous, ideal concept of "AI", but that's not what these things are. These things are fancy widgets that take an input and give an output. They are simply products that people use for the output. If they find that output useful, then great, if they don't, they won't use them. I have a widget generating a useful output for me right now. Maybe its not "writing code the way a human does", but luckily, that's not the job I hired it for. What you seem to think AI should do isn't what anyone else cares about, aside from AI researchers and maybe a few hobbyists. When everyone else in the world discusses AI, they want a widget that can do x for them. Go be mad at the marketing teams who beat everyone over the head with the term "AI" instead of "potentially useful widget".

*wooosh* good lord.