Are Large Language Models Really AI?

It ain't face pics ....

I honestly think priesthood is older.

I have tilled land with a tractor. I know how to tan hides. I have never even drawn a compound bow but I have reloaded rifle ammo.

Three words: Cabbage patch dolls.

Two words: Beanie Babies

That's something people just aren't getting cause all the hype and our human tendency to anthropomorphize everything. The real useful stuff is going to be the ability to have it search and organize your own datasets, aka agents. Having it go over your calendar and documents and act as a sort of virtual-secretary with basic task automations.

"Tell me when I watched movie X", or "what was last quarters sales numbers for product Y". "Put a reminding for next Tuesday to do thing Z". Have some sort of task or scheduling API and it can then communicate with third party services to make appointments and send communications based on what you tell it.

Beanie Babies were all identical within each series. Cabbage patch dolls were each and every one different.

huh i did not know that, another new fact from the wise leather crafting lady :)

It's a tie between a bodyguard / soldier and a "girlfriend for an hour". Proto-human male who defended nice safe cave vs female who slept with him for access to that same safe cave.

Id also say Model Context Protocols are pretty huge.

The ability to convert vague human style commands into specific arbitrary functions is beeeeeg

Right now most home assistants have to have programming by hand to handle the dozens and dozens of different ways people can ask it to turn the heat of their thermometer up, or set a reminder.

But with Model Context Protocol, an LLM can infer from their human request what they want done, and invoke the appropriate MCP

You could ask your home assistant "turn that down" and contextually it'll know you mean the music, and down, and the current volume, and thus to invoke the "SetVolume model context protocol.

Translating normal human language into executing code functions is gonna fundamentally alter the way we interface with our devices in a few years, imagine "Okay Google" but it *actually* can interpret your desired results 10000x better.

That has nothing to do with "AI", it's just the front end command parser. It extracts key verb statements from an arbitrary input string and renders them into action items that are then fed into something.

Trigger warning: this response has nothing to do with AI.

In the last couple of centuries we have run across primitive tribal societies that have no leader class or soldier class, but do have shamans. I don't know if there have been any without pair bonding.

As far as safe cave access goes many profess to see marriage as simply long term contractual whoredom. I am not one of them.

Saevel talks with a lot of authority on things he knows nothing about.

It has everything to do with "AI". Understanding human language and transforming it into commands is how Alexa works, for example and it's one of the key selling points for LLM.

Stop being an idiot Sandwich.

What Saevel is saying is that interpreting free-form human text does not require understanding it.

And he is correct. The LLM does not "understand" the text in the way humans do. It simply interprets it, picks out keywords to put in a model and then processes the model.

Much like a compiler takes programming language, a human made up language, and converts it into machine code for computers to use.

That's not what he said. Why are you sugarcoating it? This is what he said:



And he is wrong. I didn't just give my opinion. I gave an example. That's what Alexa does and Alexa is not a front-end command parser. ChatGPT, which allows you customize it to convert speech to command, is also not a front-end command parser like Saevel is saying.



Also, this is absolutely wrong. Seems like you, yourself, don't know what you are talking about it either.

Lol, ok. Whatever man.

Bet you believe the AI in Rick and Morty is real too.

That's silly, the wife isn't required to have sex any more. I was going to say that it's more like a non-compete or exclusivity arrangement, but infidelity is becoming the norm.

I guess all it really amounts to is a way for a lesser earner to socially pressure a greater earner to tie their finances together under the guise of love. Since it's 2025, that statement intentionally lacks any gendering.

I'd service a cougar in return for being cared for financially. Just sayin.

No, I actually do research in the area and publish articles on the subject. Unlike you and Saevel, I actually understand how it works. You still don't see me going ThATTZ NOtZ hOWW iiTTT WOrRks every time someone pots something.

You are more likely to believe in Rick and Morty than anyone else around here. Just carelessly throwing complex terms like "human understanding" on game forum like you understand the concept.

It's very important to separate out command text parsing from the LLM since they are entirely separate things. Those command parsers have existed for awhile and are present in many things that have nothing to do with AI, LLMs, CUDA or model parsing. People keep anthropomorphizing things and then making judgements based on that anthropomorphization.

I've mentioned it many times, LLM's are just giant multi-dimensional arrays where each dimension represents a unique value of that language and each data element represents the relationship of that value with every other value of that language.

Take the verb statements from a command parser and feed them into the LLM looking for elements with the highest probability gives you the most likely values to follow those verb statements.

He believes whatever the loudest voice in his head tell him, which only really happens if he forgets to take his meds.

This is not a joke...

I'm talking about proto-humans, which existed well over 100,000 years ago. The whole sex for resources / security exchange that forms the homo sapiens mating system has been around since before homo sapiens existed. Monogamy / structured religion are fairly new things for us. It's not discussed often because it becomes a fairly uncomfortable conversation but genetic analysis of male and female homo sapiens has the successful reproduction ratio of between 5:1 and 10:1 female to male. Essentially 80~95% of males never reproduced. Enforced monogamy seems to be how society found a way to motivate the formerly rejected males to contribute something useful.

Idiot sandwich who blocks people who refute him uses Deflectaga again.

Whatever you spilled there is just freshman throwing terms at a poster presentation.

You got refuted so, just stop pretending you know what you are talking about. CUDA, large arrays... stfu idiot sandwich, and stop correcting people. You are not that smart or knowledgeable.

I've met lots of PhD wannabes that think research papers matter, it's fine.

In the USA. In this century. For non Catholics and non orthodox Jews. (I know nothing about LDS practices in this regard.)

We can't even infer the social structures of proto humans. We know what tools they made, and how they made and used them. We are starting to understand their migrations and interbreeding, their diets, and some of their hunting and gathering strategies.

We do have primitive human cultures to observe.

This is my last reply to this particular derail.

Clearly what matters is idiots throwing terms they don't know anything about on FFXI AH. Then respond to every post with: tHATSZ noT hOW IT wOREKz ACCTUUAAALLYYYYY.

If you don't know how stupid statements like " LLM's are just giant multi-dimensional arrays where each dimension represents a unique value of that language and each data element represents the relationship of that value with every other value of that language" are, you shouldn't be talking about the subject or at least not correcting people.

You can legit google this and in less than minute you would know that the way Saevel talks is like a junkie that never made it, but likes to pretend they did. That's why he tries to be big here, not elsewhere.

Look you guys need to kiss and make up. Far too pent up.

To celebrate pride month!

Nah he just still butt hurt from the other day.

I swear half the people here think the meta in LLM is still only autoregressive models

Lmao, idiot sandwich, every single interaction we have had here I have proven you wrong. Which is why you blocked me. The thinnest skin chick.

LLMs aren't simply arrays where each dimension represents a language value. They're neural networks with learned representations that emerge from training, not predetermined mappings of language relationships. The internal representations are far more complex and dynamic than static arrays.

Also, junkie, LLM's embedding space doesn't "represent a unique value of that language". The dimensions represent learned features that capture complex linguistic and semantic patterns - they're not one-to-one mappings to language elements.

AI company files for bankruptcy after being exposed as 700 Indian engineers

No comment.

AI = Actually Indians