Are Large Language Models Really AI?

those really heavy monitors that cause cancer

Any thread that has the RAN (does not) have potential.

Critical race theory.

Not a bad explanation at all.

But CRT is a post graduate course taught in certain fields of study. The right wing freakout says it is taught at every level of schooling, even elementary, and that what it teaches is to hate white folks.

The current AI is like what we have been learning with math since highschool.

At the beginning, we didn't know squat so the teacher gave us a book with 100 problems to solve. After cracking a couple of them, we started noticing they are sort of the same thing with different variations. Then we solved them all like a machine. Then comes the next set of lessons with 100 of its own problems...

During midterm, the professor got drunk and threw in a bonus problem we haven't seen or worked on before. Most gave up but some dude will try to break the problem into smaller chunks and pieces to find if there would be something similar. This requires a certain degree of mastery of the knowledge. If the professor wasn't too drunk then they would find something similar and allow them to notice it's sort of like a combination of the previous problems and you can solve them in sequence.

So the AI nowadays can figure out something based on how much related data patterns it learned but at some point it will try to make stuff up because they look similar.

However, I remembered something I read in one of those crazy AI textbooks back in the 80s. Intelligence isn't about solving problems. It's about creating problems when none exists and trying to solve it.

Not your typical 80 YO here.

Real early on I was at the bleeding edge of microprocessors. (I have fallen WAY off it since then.) But having been on the web since dial up days I speak the language. And never having had kids I have no grandkids to run to when I have problems.

Phones... I am way behind on phones despite being an early adopter of cell phones. And if I ever buy a new car (improbable at best) I am going to need classes on how to use the ******* features.

this image in particular:


The new solutions it found were all to problems that are easily scored, which is a pretty great space for algorithmic exploration and lend themselves to AI application.

Defining and balancing those metrics for more complex problem will be the hurdle, especially in cases when the data is fairly dirty.

This is why you sometimes get 19-year-old math wiz solving world class math problems or some homeless dude solving one of the 7 millennial math problems while a bunch of other math cracks couldn't hack it with 30 years of research.

The solutions to a lot of maths are just like your average highschool trick questions but with a lot more expansive tools to work with. Normally, mathematicians would decide on the tool to use to solve and try to hack at it. Sometimes, they have to gather a bunch of tools to create a new tool to hack at it. This is why sometimes you get calculus problems that can't but solved until someone put the problem into geometry or something.

AI can go through the tools way faster than humans so they tend to find solutions faster.

Some signs of AI model collapse begin to reveal themselves

Prediction: General-purpose AI could start getting worse

Actually the article says they are getting worse.

Emphasis added.

The "The model becomes poisoned with its own projection of reality." part proves they are behaving like people. They need to model better people.

I watched a large session of Github Copilot webinars.

Last year I was already introduced to data science for an exam I partake. Few weeks ago I finished an entry level book of programming with AI.

Idk, I feel like AI fields has two interfaces: one publicly available for consumption, and the one that plays like a perfomance competition.

While trying to compete for performance is quite intricate and complex field, on the other hand the AI for consumption is quite uninspiring as it never felt like a technical or algoritmical procedure: prompt engineering feels so cheap, those webinars had total of 12h of content, and Im not sure what was added above the entry level book I had.

That means: learning how to use prompt engineering is like using a couple of “advices” to generate a good answer. I dont feel like anyone at this point in time can provide the “optimal” prompt, that could return the “best” answer with the minimal amount of commands.

That is why I dont particularly feel invested in learning more AI for consumption. And AI for competition… nah, not my thing.

Im a creative person, not a competitive one

AI is rotting your brain and making you stupid

I am positive that in ancient Greece of 600 BCE people railed against this new writing technology thing saying that it would erode out memories and make us stupid.

The Iliad text dates from 800 BCE so probably not.

Vibe coding, (Vulnerability as a Service (VaaS).

How to ask Copilot to give me the most likely answer from Stack Overflow or some guys random tutorial blog.

i love it

Ai won't make people dumber, it will ruin future generations who grow up with it though.

They won't learn to do all the things in the way we did, so even though we have all these tools today we still tend to do it without them most of the time (or at least we can). We don't ask chatgpt to write out every post we make for instance.

Children born from now on won't be like us, they will grow up in a world where the "AI" will do almost everything for them and there will be no stigma using it for everything.

Kids born from 2025> will be arguing / talking with each other with chatgpt written posts / texts, showing off pictures they made with AI (and nobody will complain it's AI, and they will get praise for "making this great masterpiece"), same for music, same for pretty much anything you can think of.

You don't see the real impact of new tech till the next generation grows up fully embracing it.

I see Generationism here, lol.

My parents can't use mobile phones as well as I can, and everytime when they ask me very simple questions about using their phone my reaction was often "you don't even know THAT? Why can't you learn something so easy?"

30 years later, our grandchild will look at us the same way. Except at that time the world belongs to them, not us. So whatever we think about them, it doesn't matter.

I don't really see a future where people 'will get praise for making this great masterpiece' when they create AI generated garbage. Seems more likely the most capable/developed models will end up privately used to create content. Corporate consolidation and job erosion.

However, I still think a lot of people are being sold a bill of goods by clever marketing and paid advertising. Right now, AI is just a combination of a search engine and a word calculator. Recent strides have revolved around plugging it's own results back into it and combining them with command interfacing to create 'agents'. The core model hasn't evolved and the more times you have to feed back through that model, the more unreliable it gets. So, attempting to increase accuracy by using agents to search and requery results in far more hallucinations and errors. You can't build AGI on top of it; the people working on it know this.

I'm not saying there won't be an improvement that makes it possible, but I find it extremely unlikely anything approaching AGI will ever come out of LLMs. It's the wrong model for the problem.

That is the oldest job in history.

Id say the word "change" is more appropriate

"Ruin" asserts current culture is superior, but the reality of the situation is, I doubt literally anyone here posting knows how to til land, tan a hide, hell I bet maybe only 1-2 of us can even string a full size compound bow

If you told that to someone 500 years ago, theyd probably say our generation is "ruined", how can you get by in day to day life if you can't even string a goddamn bow?!

Well, the answer is, no one gives a ***lol. It's become a completely irrelevant skill.

Instead, chances are culture will evolve to focus on other arbitrary stuff, and that stuff will become what is actually of value.

I strongly expect that handmade items, bespoke stuff, will become the primary reflection of wealth. Purposefully imperfect items, masterfully crafted to be intentionally human.

Getting AI made stuff will be easy, but a lot of the same.

We already are on this route, you can get mass produced garbage from china for cheap, but wealthy people get their ***made by hand.

It's gonna be that but multiplied a hundredfold. Everyday items will be dirt cheap, mass produced with little to no humans involved, but items made by hand will become all that people care about.