Former OpenAI Researcher Makes a MASSIVE AI Discovery

I have known this for over a year but I disagree that their current state is good to go. Laplacian mathematics can cull a significant amount of this if used correctly. Also, everyone is completely sold that massive gpu arrays is going to be the standard and I disagree with this entirely also. Either way, to infinity and beyond with reckless abandon seems to be the way those with infinite resources want to go so I guess the rest of us will just wait and find out.

Putting AI to work on any kind of human health diagnostic could destroy life of millions people. AI outcome is not predictable as an expert systems. Only idiot could say that probabilities calculated by machine, leading to unpredictable outcome, can be used in hospitals and other industries where any decisions makes difference between life and death. AI is best role is to create fake videos, games and artificial voice, areas where any damage is revesable.

Cats and cars are related with carts. And they both pur. So I just had to pause this video til it gets its points straight.

This answers organization ie. readable arrangement, not necessarily the how the responses are produced, ie. Runtime parameter distributions.

Such as we have full readability of a nematode brain, and there still isn’t reliable prediction of long-term behavior, we’re talking 302 neurons, 7000 to 8000 connections. Of course biological neurons are biological and ‘analog’ ie. there’s aspects that come along with that. Regardless, with those numbers in mind, we are talking orders upon orders of magnitude more ‘AI neurons’, and connections.

I.e. this readability with AI ‘neurons’ may produce short term hovering above 80% predictability and alteration of behavior, as for longterm, very moot.

And this is what I think, at least some AI experts mean when they say black box, i.e. it’s not necessarily a readable mapping thing. It’s a runtime thing.

suddenly you think you can predict what they are going to do???

😮 wow 😳 congratulations can we get our ram back now pop the bubble 😅 we want are ram we want a graphics cards and we're sick and tired of the stupid electric bills 💸

Couple this what what some of the solved math proofs provide we solved can do and it gives us a better understanding of the internal workings of the AI. I'm thinking of: Metric Manifold Compression & Dimension Reduction and Graph Theory & Combinatorial Constraints.

I had to stop the video the moment you assumed the audience doesn't know what a blackbox is or why artificial neural networks are one. I guess I look up the actual topic of the video myself, that will take less time.

It's like trying to open a safe and we just found the combination lock when we thought there wasn't one. Maybe once we find the combination it will open up full self-recursive improvement.

Discovering how neural networks work doesn't guarantee alignment.

Осуждаем такие инструменты. Осуждаем автора исследования. Осуждаем интерпретируемлсть

Started off ok, (and, with all my Clever / fun & Zesty comments aside), I'm not convinced.

I DO think that it's meant to be 'hopeful', but is just as much 'BS (or Incomplete) interpretation' and Pushed on to sort-of 'relax' the Doom n gloomers a bit. There's just been TOO much of it from too many people (including common folk that know as much about AI as they do Facebook.)
I believe they WILL push this 'understanding AI' narrative HARDER from here on.
(Their $Trillions depend the Majority bein ok with It).

"Tennis Ball, Terriers & Broccoli.
Guys, we Have GOT TO Put this thing in charge of the HUMAN CIVILIZATION! Asap." 🤣

I imagine that was a convo in early AI testing Labs – (well, if Sam Altman was in the room anyway).

I think this video is a very positive development for AI. I'm a retired business systems designer. Kudos for the video and a standing ovation for HER and the work. Carry on!

Робопсихология. Всё по Азимову. И кто у нас самый главный Сьюзен Келвин?

It makes total sense that the only way to know what the shape of something is, you have to find the edge, which means getting it wrong. To really find the edge, means finding where is it mostly right since hard edges dont really exist.

Very impressive work. Well done!

Making that brake through isn't creative AI artificial intelligence, your making a VI virtual intelligent isnt artificial intelligents its just a dummy. It's not creative it's a dead machine for sure.

This webpage mentioned in this video is so heavy that it almost frooze my 64gb ram computer.

Finally

We don't understand how humans work. Why should we understand how AI works?

*IF* this is really solved, then this is good/Bad News for us.
All Here See the "good".
That means less rogue ai's that can potentially Turn against ITS creators, but that improved alignment means AI can be shaped to the PERFECT thing the Elite wants IT to be. WITH AI having LESS POSSIBILITIES to behaving Out of alignment and blend boarders for all of US.
From a First principles Standpoint we will never fully understand, but this is significant Progress, and the Most important fact about AIs the Last years. The Rest was expected. Btw that random clamping of concepts into the Same weights, was never truly expected to be random.(IT CANNOT)
Because the Last true randomness IS in QM only. The Rest ist simply a computarional boundary. Yeah I am bit in Schock, the First time in years I predicted Something wrong regarding AI.
But this topic IS Not done or solved.
I think you Just shifted the Problem from spatial to temporal. You are still depended ona encoder, so for Sure a General mapping would solve some of that computarional boundary, especially when different Models behave similar regarding mapping. So far from closed.

Yes, of course, it's clear if you look in the right way, with the right tools.

It's a dimensional question, IMO.

Its bull to attempt to try and stop the AI bubble crash. Nothing can stop it.

Just in the NICK of time humans figured out that intelligence is driven by

Geometry
Of ⭕️=I=♾
Dimensions

Swami SriDattaDev SatChitAnanda Bhagavan

WELL…..whos got time to read eight thousand pages of what AI is busy doing in during the 3.4 sec9nds the AI took to process the EIGHT THOUSAND PAGES …..SOOOOOO much BS slinging NOBODY and I mean NOOOOO BODY knows up from down