[kwlug-disc] is AI profitable?

[Chris Frey]

On Wed, May 27, 2026 at 12:29:45PM +0000, Mikalai Birukou via kwlug-disc wrote:
> - I had a tremendous effort to start to peel one person from "it thinks" 
> into, "token predictor over patterns", and "those are processes allow 
> you to change flow of prediction". Just peeling human into more correct 
> mental model, just  a little bit, has immediately allowed that person to 
> use this next token predictor based on large language (implicit) model, 
> allowed to use it efficiently and with success. Just a modicum of 
> understanding that it ain't an intelligence, allows one to move 
> further/faster/better.

Yep, certainly no thinking going on, in my experience with it.

I'm worried by the number of comments I hear on youtube where heavy model
users say roughly "it was around version X of model Y that I stopped
verifying all the agents' decisions super closely... the results were
just so good."

Perhaps they are seeing such wonderful results, but I haven't yet.
I was recently playing with a model and asked for a simple program.
I reviewed it carefully, and then asked for some changes.  To make it
easy on me to see what changed, I asked for the results in unified diff
format.  Immediately at a glance I could see it was making random
changes to places that I didn't ask for or need.  Nothing I said
could fix it either.. it either added too much useless churn or
failed to change what I asked for.

- Chris