Hmm the laws of physics work a certain way. AGI works a certain way, a way like nature's laws but also in a way that manipulates it.
If it can think in video, it can think about and with physics, to research AGI no? But then text is efficient and uses pure information in its smallest form oh what am I smoking?? Is not AGI just a crazy match system that tries to recognize unseen problems by trying to match it to many already experienced similar old past similar problems it knows the answer to, hence how it "discovers" so fast the answer from seemingly thin air. Is not AGI just doing hello=hello in a full scale assult? I mean hellllllllo=hello, Heeeeekgmgllo = hello.......olleh = hello.......................................................................think about it guys, if hell (no o yet) is the most easiest input to match to hello, then it predicts next is the o to the input, and if hEll is very similar to the memory hello, it matches a bit less due to cap letter, and so what I'm saying is the more you distort hell as input the more possible types of it in more less similar amounts there will be, until you get trillions of types of 'hell' that are each very barely likely more to match hello in the brain, until eventually all inputs that are random would match neutral to hello. What I'm SAYING, is that AGI is just a exact match system hell=hellO pop out the prediction 'O' = hello BUT what I'm saying is that AGI is the goop that falls of Terminator and latches on like alien tech to adapt to the host, it is able to see how something, of many cases, while barely similar, matches still by how much amount the word 'hello'! In saying that, if AGI is a match system like so, combining millions of matches similar memories, be it translated versions of the word hello or similar words like welcome, or visually the hello act......then maybe to solve AGI is the same thing we've been doing but just have to DO it more, like Moore's Law. We just keep doubling, we just keep matching BETTER, more flexible, filling up more holes in the road. Yes there is exponential curves in a neural network n stuff like rewards and how many time recently seen making it predict it lots ex. eeeeeeeee so you likely should predict e is next yes, until forcingly forgets it thinking it was done said now you say next word......thank you so....you're only going to say 'much' now see? As everything has an end, eeeee will too, and has a known average time it ends too if seen lots. Next token prediction also works same way, its like matches all instances of hell>? in a book and seeing 'o' is usually comes next. Many matches, and out comes intelligence. That is just exact match, as I said there is more such as H3LLLLLL> which will match hello in brain and trigger the o at the end to be said. Yes AI ANNs seem to be similar to my AI master theory BUT seem to kinda just gradient descent the friking thing and just learn all the connections on tis own, and how to average them together and stuff I THINK am I wrong um??? But, this is still the same thing really! It uses embeds, look at that! Etc. Aside from memory combination and some stuff, it is similar. But I wanted to say it is like Moore's Law, the gradient decent learning in today's ANNs is learning more like that self latching-on goop that is adaptive. It might do some the stuff on its own! Such as eeeeeeeee>e comes next energization, etc. I don't know about that one though but just giving example. Anyway, very interesting thoughts I share here maybe? Hmm back to the question, do we need video AI? Well, AGI is a good exact matcher if you get me :):), so if it is going to think about how to improve AGI in video format style, it would dream of the matching happening. But it can too in text form hmm. Maybe video AGI is the only way ahead to manipulate physics SO THAT the matching can be done the best and most scaled up efficient, doing more matches with less resources, like the Moore's Law thing I said going on but for matching more and more utilizing all memories more than otherwise. *Because text can't create and test that chip brain, video AI can. You need to create it. A text AI can think about it, but it can't say ok let's write the code and run it and......it can improve its code but NOT the hardware it runs on !!!! So video AI can do that, fast ! Computers can simulate real world tests FAST, hardware development will be so fast then soon.* Code can do anything, really, but it can work faster if the hardware is better! ? ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5ac1f9ef84312f96-Mc80fa22d4a11f7599d7d1e6c Delivery options: https://agi.topicbox.com/groups/agi/subscription