Not remotely impressed with AI

There's a good article in the Wall Street Journal this morning that describes research showing roughly what I have concluded by amateurish intuition - put politely, current AI is nothing more than vast memorization of innumerable rules of thumb, plus the ability rapidly try out those rules and discard errors.  

Less politely, but very funny, is the description:  "All of this work suggests that under the hood, today’s AIs are overly complicated, patched-together Rube Goldberg machines full of ad-hoc solutions for answering our prompts"

That perfectly describes a garbage in, garbage out tool.  More alarmingly for those investing in AI ventures, it succinctly points out a structural design flaw that will eventually prove fatal to these projects - it's roughly like a friend of mine who once excitedly told me that they could now 3d print wood.  I was naturally skeptical and started asking for details of how this was done and quickly pointed out that what he was describing was not wood - it was a plastic like material using wood as a precursor material.  Similarly, there is nothing intelligent about artificial intelligence, we are simply calling it that.

I'll post the link, but it is likely paywalled.
https://www.wsj.com/tech/ai/how-ai-thinks-356969f8

Amen to all of the above.
  Ron

With all the hype about AI (A1 according to some elites in the government), I can't understand why it's not actually used effectively by businesses. Amazon customer chat is clearly AI that cannot move off a set group of responses and will endlessly repeat steps until you quit trying. Amazon search is likewise unable to narrow it's results, even when you select one of it's suggested search terms. Finally, my credit card company endlessly suggests I get their app for better service when I clearly have, and use, their app. (Even my medical practice group is touting it's AI capabilities. Maybe I won't have to wait months for appointments anymore, assuming they have adequate bandwidth.)

I wouldn't bet against the future impact of AI. First, it's only in the learning stage at this point. Think of it as unlimited relational databases being built that can be accessed and data recovered in almost instantaneous time. The analogy of GIGO (garbage in, garbage out) is accurate at this point. The real benefit will come when they develop the inference capability which you all perceive it's doing now, but it's really not. That technology is only now being developed and deployed which will unleash the real power. Right now I use ChatGPT to research things that are documented, and that I can verify, but would take hours to find on my own. A recent example was looking for some genealogical historical information about the areas that my ancestors settled nearly 400 years ago.

Dale Stansberry wrote:

"The real benefit will come when they develop the inference capability which you all perceive it's doing now, but it's really not."

We don't perceive that, we are told that by the hyping hucksters in the AI business.  Some of us believe it, some of us do not.
I don't believe the current LLM paradigm will ever develop inference capability.  To the limited extent that I understand quantum computing I would say maybe it could.  Maybe.  Burn that bridge when we come to it I guess.

Peter Martin wrote:

https://www.woodcentral.com/?p=WAGI

I like it!

Seems to have a fixation with serial killers!

My girlfriend would definitely agree on the irresistible sawdust cologne.

Another interesting article:
https://www.wsj.com/arts-culture/film/tilly-norwood-ai-actress-particle6-d5c51da9

I'm still not impressed however.  Much of the reason for the lack of being impressed is that although the tech is genuinely astounding, the "intelligence" is not.  It took months of tweaking - the exercise of human judgement - to refine it to the point where it's marketable.  And still the AI kept going back to those bushy eyebrows mentioned a couple times in the article.

This is the sort of thing that makes me doubt very much that the current LLM version of AI will ever get to the inference stage.  It's throwing stuff at the wall to see what sticks - with a huge database available for making informed guesses as to what might stick best.