Perhaps the BEST B1G Forum anywhere, here at College Football Fan Site, CFB51!!!

 *lots of good stuff*

One of the things that has changed is that now older AI's are taking an active role in developing newer AI's.  This has been done previously, and talked about at length with lots of projections and speculations.  I think the exponential curve of development has been understood to be a possibility for a while, and even expected.  But drawing on what you wrote, it kinda seems like now is the first time we're really starting to see it in reality.

And it's kind of like the Grand Canyon.  You can know you're going and what you'll see.  You might have seen pictures and had people tell you in detail all about it.  But when you actually get there for yourself, no matter how much you perceived it in other ways, it will take your breath away.  

the 'real', in my humble opinion, trick that makes AI work happened back around 2004 or so... some guys got together and created a means to create databases on the fly... eliminating the need for tables and queries.  those were prompted by the 'user' and in real time a database was tossed together.  

it was at this point someone realized those mountains of data being collected actually had value.  they didn't have to be cataloged and placed in rigid places that were relevant to their use/context and then cross-referenced or existing in another place altogether in another context- they could be scanned and located just once.  The adjacent words provide the relevance, pretty much just like in 'context' of writing or speaking.  So using this approach the piles and piles of data could simply rest instead of being shuffled within a massive sprawling complicated overwhelming to most computers database.... a series of tables was constructed real time and then it was tossed when done. 

maybe some of y'all recall, but i tried to install a function here, and i still have it somewhere, that makes certain words hyperlinks and to whatever item or information they were referencing.  i got the idea from an advertisement company that did almost the same- but the links (you'd get a little tool-tip over the hyperlink when you hovered with a functional screenshot of the linked page) would go to whoever the company they were contracted to advertise went to.... i tried to make this happen to feed a search function which rewarded a thread or post based on simple quantity of times a certain phrase was used.  What was discovered is the function was incredibly resource hungry and pages were disrupted with tons of little links with tool-tips... which made the page crawl, and, I couldn't seem to control how many times it would try to run- any time a mouse jiggled or a scroll happened or input into a form field was made it tried to run.  so... it hit the circular.  I mention it because- the results (not the means) isn't far from what this thing I'm speaking about does... it collects every datum and discovers it's place in context, and looks for pages/products relevant to it before shoving it through a filter and sharing it.  

imagine piles and piles and piles of piles of data collected from phones, personal computers, servers, websites, et al.  and all of the sudden being able to use it in a functional way- not just a statistical way.  the information available is at the edge of not being able to conceptualize by most people (by most people I likely mean 'me').  it is shocking, to me. 

from that came probabilities. it may have been the first version of AI.  it rapidly collected any information regarding whatever the prompt was and gave probabilities.. it spoke akin to the old google- "87.3% relative" as an example... those were quickly translated to literal probabilities- and the user of this mechanism could offer- based on presentable evidence *someone may or may not have been able to collect themselves- what the likelihood of something happening, being successful, or failing, would be.    

(* the creators and users of this thing, at first though the system had a bug when it would collect items that seemed not relative to the task, but learned that the system was capable of seeing trends and relationships they weren't- it was a 'bug' in the mind, not the machine, so to speak.)

so as the tale continues, and interjecting some wild information here:

several gov't organizations had keen interest in this.  they contracted its development to a couple companies that were literal recent start-up nobodies.  they were smart in the fact they hired the best defense contract writers- and they pulled the wool over uncle sam.  the contracts were arranged in such a way the information requested belonged to the gov't, but the mechanism didn't.  it remained property of the companies.  even though the gov't paid for the development of the system it didn't belong to them... just what came out of it and only what they requested come out of it.  the companies knew the value of this thing to advertisers. they can granularly target YOU with advertisement and have a one-shot mechanism for companies seeking to advertise.  

at any rate and cutting this off, the device described has evolved.  data can be parsed faster; it uses phrase collection and reconciles against adjacent word use which is two distinct functions happening concurrently, and now, it 'remembers' useful functions and stores them away for future usage.... almost like a script cache stores script functions so it only has to run once- logging the result and retreading it as needed, and finding new relationships along the way.  

AI isn't sentient or even singular... there is debate if that will ever be done... it's artificial.  it mimics, and that is all it'll ever be.  as it's access to data that replicates personalities is indexed over and over again, and new arrives every second (those phone carriers? they found a place to dump the data for profit- something that was a flush problem before) and makes it's rendition of what it thinks we are more and more convincing... but it can't ''think', it can just compare.  hell, there are some people like that and they certainly have their use, but..... whatever leap (and there will be one) that happens technologically has a ceiling if/when AI takes over.  it may take 100 years to get there, and it'll certainly be beyond whatever we could do on our own without that function, but it will be stymied by past.... without human interaction. ... like a great big race to get to the 23rd Century, and then? We stagnate or maybe even just disappear as a species.  

re: six fingers and seven toes.... AI is now tagging images made by AI so AI doesn't replicate or process information that is wrong without knowing it 'could be a bad source of information'.   that's kinda funny, no?     

I would be fascinated were I working now to attempt to use AI in my work.  Maybe we could ask it who will win the NC next season ....

I glanced over an article about someone using AI to select stocks, maybe a mutual fund?  It wasn't going well.

We use AI to forecast future demand directly, and to develop additional forecasting models to forecast future demand directly.

In both cases my own forecast beats the AI models by 20% or more on average.

On the models that I've taken time to train directly, I'm still beating them by 15% or more on average.

But I'll tell you this-- 5 years from now the AIs are going to be better than me.

AI is like a psychologist that will tell you "the best predictor of future behavior is past behavior.".... and it's true... except for those wild cards that... well... are wild cards... they happen with such low frequency or in such timing or space they are hard to account for.... like cards being shuffled or somebody calling in sick at a job and the butterfly impact cascading downline.. AI would regard that as a statistical anomaly and be right.  but us?  we gauge everything off of a different way of processing that same dataset. rationale mixing with emotional response- or gut feelings that are really just your sub-conscious cutting to the chase.