Subj : Re: AI is sucking up powe
To   : AARON THOMAS
From : Mike Powell
Date : Fri Jan 03 2025 10:20:00

>  MP> I am having a discussion in DoveNet TechTalk where the other party
>  MP> brought up how AI has sort of been used for years in their field.  It
>  MP> provided various mathematical and data point information that was
>  MP> accurate.  It is when they try to get AI to do things that require it to
>  MP> "think like a human," rather than a computer, is where they are getting
>  MP> into trouble.

>Into trouble with electric consumption? (Or into trouble with the
rogramming?)

Into trouble with the actual results being incorrect.

>  MP> Webservers comsume electric for sure, but they don't do a lot of
>  MP> processor intense work.  The processors in an AI datacenter are going to
>  MP> be different -- think more "heavy duty" -- and the machines are going to
>  MP> pull a lot more power than a machine that serves up web pages.

> I appreciate your explanation. I just don't understand what constitutes "AI"
> though, and I don't understand why they would need (for example) more cores,
> more RAM, or more bandwidth than an average webserver would.

Because the processes behind it require a lot more "thought" by the
computer, for lack of a better term.  Serving up a web page requires little
thought.  Making decisions that a human brain might make requires a lot more
thought.

In comparison, if you are doing a job that requires you to pick something
up out of one pile and put it in another, that will not be very mentally
taxing.  OTOH, if you have to pick up something in one pile and then decide
which pile it goes in, that requires a little more brain power.

Doing the above while keeping count of how many you put in each pile, and
knowing to send the pile along and start a new one whenever that pile
reaches a count of 20 requires a little more brain power still.

The more complex the task the AI is asked to complete, the more "brain
power" it will take.  So, it needs more complicated, more advanced, and
likely more power hungry CPUs, GPUs, etc.

> The phrase "AI" is ambiguous. I did a Google search of "Which language is AI
> written in" and it said "Python," but is that accurate? It seems like there
r
> better programming languages to use for tasks like emulating human thoughts.

I would assume it could be programmed in any language.  Python is one of
the hot languages these days.  Rust is another one.  I have no idea which
would be better to code an AI application in.

> I get what you're saying, and it's a good guess, but to get to the bottom of
> this I'll have to (someday) learn why AI programs are more resource dependent
> than others.

It would be the more complex decisions they are asked to make.


 * SLMR 2.1a * "I never met a chocolate I didn't like." --Deanna Troi
--- SBBSecho 3.20-Linux
 * Origin: capitolcityonline.net * Telnet/SSH:2022/HTTP (1:2320/105)

.