XPost: alt.folklore.computers   
   From: antispam@fricas.org   
      
   In alt.folklore.computers Stéphane CARPENTIER  wrote:   
   > Le 11-01-2026, John Ames  a écrit :   
   >> On Sun, 11 Jan 2026 01:52:02 -0500   
   >> c186282  wrote:   
   >>   
   >>> Look ... nobody is going to be 'writing' much of ANYTHING within five   
   >>> years. The "AI" will do it all - probably led by the pointy-haired   
   >>> bosses who can't find their ass even with a spy sat.   
   >>   
   >> The "AI" bubble isn't going to *last* another five years, full stop.   
   >> Frankly, I'll be shocked if it makes it to '28, if that.   
   >   
   > The AI bubble which must be considered (I mean: a lot of people has a   
   > lot of understanding of that term) is that big companies will stop   
   > to invest always more money on it. That doesn't mean the data centers   
   > will stop to work, it means that new data centers will stop to be build.   
   > At least at such increasing speed. It doesn't mean that actual GPU will   
   > cease to work. It means that big companies will stop to buy so many GPU.   
   >   
   > So, globally, everything that has already be done will stay. And new   
   > things will improve at a slower pace. The AI is there and will stay   
   > there for a long time. When the AI bubble will burst, its impact will   
   > be more on the global economy than on its usage.   
   >   
   > That's why the companies invest so much. Because the one which will   
   > have the lead at the time of the burst expects to keep that lead for a   
   > long time. Not because they are stupid and can't predict the obvious.   
   >   
   > The AI is there, like it or not, you have to live with it. The fact that   
   > you or I like it or not is irrelevant. Like when Platon was criticizing   
   > writing system because people stopped to learn by heart, the writing   
   > system was there, stay through the ages and revolutionised things. There   
   > are some things like farming, writing, electricity that changed   
   > everything on the human way of life. And the AI is one one them. There   
   > is no going back. I'm not saying that it's good or bad, I'm saying that   
   > it's the evolution (not progress because progress is good by definition)   
   > and one can't do anything but live with it.   
      
   I think you miss the point.  First, a lot was promised and there   
   is still hope that big thing will be delivered.  So, active players   
   may win really big, that is why money still flows in.   
      
   Now, concerning burst: AFAIK AI companies use investment money to   
   cover cost of operation (whole or in significant part).  If there   
   is burst, they will have to stop operating literally closing   
   their datacenters.  Basically only things that generate profits   
   or possibly some research by companies that have other sources of   
   income and still want to continue research.  But that would be   
   at much lower scale than currently.   
      
   Now, concerning 'AI is there', there was significant progress in   
   some areas like machine translation.  "Creative writers" may be   
   concerned.  But there were attempts to replace a lot of professionals,   
   notably programmers.  Examples indicate that "AI" can create   
   small, trival pieces of code but does not really work for   
   bigger and more complex things.  To be useful for programming "AI"   
   and way it is used must be significantly improved.  It is possible   
   that slower, gradual improvement will lead to "useful AI".   
   But it is also possible that alternative approaches, currently   
   underfunded due to AI race, will progress and be used insted   
   of "AI".   
      
   Let me say that I am an AI fan and that eventually we will have   
   useful AI.  But current trend seem to be fundamentally wrong.   
   I mean, educated young person got maybe 20 years exposure to   
   learning materials.  Assuming reading 10 hours daily at 10 cps   
   we get about 2.6 GB of learning material.  Beside reading there   
   is speach, smell, touch and visual innformation.  Speach has   
   similar speed as reading, smell and touch seem to be lower   
   bandwidth.  Theoretially visual tract has huge bandwidth, but   
   blind people seem to be as inteligent as others, so visual   
   part is likely to be noncritcal for general inteligence (as   
   oposed to some sepcific visial tasks).  So 2.6 GB looks like   
   very generous estimate on amount of training material needed   
   for human-level inteligence.  Yet current AI uses vastly bigger   
   training sets giving much lower performance.  I have heard   
   from neural network specialist that current network training   
   algorithms are vastly more efficient than traing going on   
   in brain (or rather vastly more efficient than our current   
   idea of what is going on inside brain).  Yet computational   
   cost of AI training seem to approach estimated peak compute   
   power of human brain times time needed for learning.   
      
   So, it looks that for general AI we are missing something   
   important.  For applications ANN apparently struggle with   
   tasks that have easy algorithmic solution.  So natural way   
   forward with applications seem to be via hybrid approaches.   
   But AI crowd seem to prefer pure ANN solutions and tries   
   to brute-force problems using more compute power.   
      
   --   
                                 Waldek Hebisch   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)