About "AGI is Here. Anthropic Just Proved It". Honest opinion

Hey everybody.

Pepe here saying hello.

I am very surprised you just posted "AGI is here". REALLY, VERY SURPRISED. Honestly.

Let's talk about something. Currently LLM's, THE BASE OF AI (I am really starting to hate the term AI because in my opinion is not Intelligence, more of a marketing term) is: Neural network, Transformer type mixed with MoE (Mixture of Experts) with a lot od different forms and types, but that is pretty much it.

Let me say this again: LLM's are neural networks. This is old stuff. It is amazing how LLM's can predict a lot of things, it truelly is, and it is a spectacular tool for developers. BUT IT IS STILL A NEURAL NETWORK. It is rigid, and it has to do a lot of stuff just to get the context right, and with this in mind it still halucinates.

Let's talk about AGI for a second. For AGI to appear, i think it really has to improve Retreival of information. Humans can listen 2 seconds of a song and evoque a memory that passed 40 years ago, and also do the same with new stuff. A brain can make new conections, make reinforcements, correct itself with a little amount of energy.

@Nate Herk

do you really think LLM's right now are as maleable as a human brain? I am sure they do not work remotely close as a human or even animal brain. We are far from AGI. We can make really good experts and solve a lot of things, but for it to be general, i don't know. It is arguable that maybe LLM's doesn't need to work as a human brain. I mean airplanes do now work as a bird does, but in my opinion, and i think a lot of us will agree, on biology and evolution, we always tend to imitate animal like patterns in order to do things well.

With this said: I think AGI could do something different as a brain does (like an airplane does not have to move their wings in order to fly), but still LLM's right now have a lot of difficulties on retreival of information, and learning new things. LLM's right now are rigid, not near what a living brain of any kind of creature does. So how would LLM's solve the pattern that makes a human think like a human? It cannot be done, it'd have to make "maleable LLMs matrixes" or dynamic matrixes that learn from new context and that is not possible right now. (I agree that MoE is perfecting as the day passes, but is this enough to make AGI? I don't think so)

So Nate, could you define us what do you mean about AGI IS HERE? Could you define AGI? Because, right now it's ALL ABOUT CONTEXT, and Mythos or any other right now are related to the discovery of Transformer Neural Networks (i think this happened 2017) and neural networks that are known 70 years ago.

Let me know what you think guys.

Peace

2 comments

About "AGI is Here. Anthropic Just Proved It". Honest opinion