Are Large Language Models Really AI?

That part too.

If we're talking the official ones, it was originally the ability to mentally process information. They later ... tweaked it to include stuff that has nothing to do with mental processing, but that's a different discussion. The entire thing is timed and attempts to see how quickly and accurately the person can find patterns, infer information and extrapolate on those patterns and information. Basically how fast can you solve mental puzzles.

Having said that, as I mentioned earlier "AI" is nothing remotely close to "intelligent", "sentient" or "learning". Nothing but statistics expressed through the English language. It's a google search on steroids.

We as humans have this tendency to project our anthropic identity onto anything that even remotely resembles us. We do it to animals, planets and even natural phenomena. Right now people are merely projecting their human identity onto a mathematical construct.

Kyle Hill did an extremely good breakdown on what Chat-GPT and "AI" in general is without using pseudo-magic click bait words.

What's your comment on it being able to reason about math? I also wonder if our brain is actually pretty close to a statistical model, just like LLMs.

Our brain is very very different, nothing remotely like that.

LLM's are stochastic parrots, they are just repeating back what was fed into them. You put in an input string, it breaks that string into indexed phrases and then integrates through the array to find the most probably match for what follows. The framework they use was designed to predict patterns in very large sets of unstructured data. The only thing that makes "AI" different is the dataset being all the public and not-so-public works available off the internet. It's simply finding patterns in the human language and repeating those back to you.

The only part that is even remotely "human like" is that the training algorithms were based on an old theory (since disproven) of how humans learn. The math's and theory for these models have been around for decades, there simply wasn't a feasible way to crunch the quantity of raw data that is required to train the model. It would of taken a super computer a very long time to do previously, then CUDA was made and someone figured out a way to adapt the algorithms to work with that instead.

Everything else with "AI" is just buzzwords and smoke and mirrors used to hype something and make people rich.

Thing that annoys me is the sheer amount of gaslighting going on (mainly from S Altman), to try push up stock prices. Everything from blatant fear mongering to embellishment.

AI is going to move into two camps going forward, bad AI that is made for all the most negative reasons for shareholder/investor benefit (aka replacing workers etc) and governments (watching everything you do) and open source AI that is made to improve the tech.

Most of the advancement will be in the open source camp (which seems to be heavily skewed to China sadly), the public companies are just going to chase all the things the evil things billionaires want it for.

Wechsler-Bellevue stopped being used in 1955. I know you're old but you can't be that old. It requires reading. Standford-Binet requires reading and writing too.

You don't have to talk in any I know of. You're being extremely pedantic here - not that I was even saying reading=high IQ, but if you can't read in a developed economy there is an extremely high correlation to having a low IQ.

I think the interesting thing here is that nothing anyone is doing can not be copied by competitors - you can't patent training methods or prove how a model has been trained, and Deepseek are open sourcing all their advancements which make them better than OpenAI and all other western companies in a way - and Microsoft released a CCP propaganda free version of Deepseek already.

Qwen are the best company out there at the moment, and I have huge respect for Mistral. It's Zuckerberg that we need to watch out for the most, that man is the biggest danger to humanity since hitler.

I just turned 81.

Yes, THAT old. 1944 model. I was 11 in 1955. And despite my cancer I expect to get LOTS older. My favorite great aunt, who was born in the 1880s, made it to 104.

You can slap pat. pending on anything even if you have yet to file for a patent. Also some authors have proven that some LLMs have been trained on their copyrighted materials.

!!!

I can think of a few other candidates that are right up there with Zuck.

Am I crazy or did the title change?

Edit: Yes

I didn't change it.

Yes, you are crazy.

No.

Not because of title change

If you look at it upside down, then you are barely legal!

I'm not talking about materials but methods.

People HAVE patented methodology.

The patent office got really wonky in the 1990s. People were patenting weird things. Not items, ideas, even common algorithms currently in use, got patented.

Pretty much if you can describe something you can file for a patentent on it and in the '90s they did and got the patents.

The patent office did get better. I think.

I don't think any training methods can be patented, at least it isn't happening. This might just be the nature of the beast, that things are moving faster than even patent pending would be granted but I'm not sure. Either way, thus far, nothing stops everyone stealing everyone's approaches to pre and post training of models. MoE is becoming standard when Mistral were first to demonstrate it, reasoning approaches are standard for all models going forwards, etc.

They did. It was.

I don't know how old you are but in the '90s news had many stories on how many totally ridiculous things were getting patented.

The blame was placed on people patenting algorithms. This was a brand new and explosively broad field for the PO. Suddenly things long considered unpatentable were getting through.

At the time I was saying Costco should patent its membership store model. They could have gotten away with it.

Wait... does that mean the title DID change?

Because I swear it just said LLM models. Now it's spelled out.

I cannot confirm nor deny.

Is it a red flag to think I’m always the sane on in relationships ?

x post

Will be interesting to see how R2 compares to this when launched.

Yes.