[COFF] A thing I did (AI Slop Alert)

[Adam Thornton via COFF]

I'm pretty sure my boss gave me the task of putting AI features into our
custom JupyterLab container precisely because I was the most AI-skeptical
of anyone on our team.

Along the way, I did learn that Claude does pretty decent code reviews and
it's a good idea to let Claude do that before I inflict my code on actual
co-workers for review.  And of course I'm pretty pleased with the outcome
of the Forth interpreter.  (I did follow the earlier links to find
FIG-Forth as a standalone bootable environment for the PDP-11.)

If I have a bunch of credits left over at the end of the month I'm probably
going to see how far I get with a tn3270e client, because x3270 is pretty
long in the tooth and XQuartz is getting more and more of a PITA, and while
zti/tnz are great, I'd really like something that runs as a nice MacOS
graphical app (I don't care if it's actually Electron or whatever under the
hood), because Carbon support didn't make it to 64 bits so Peter De Camillo
decided not to try to update tn3270.  I know there's Perry Werneck's pw3270
but I have never actually gotten it to build.

The nice thing about that is that there are a series of incremental steps I
can take, each of which has a nice clear deliverable (like, step 1 is a
basic 3270-not-e 80x24 monochrome terminal), and I can do the same thing I
did with v7 and just set up Hercules running TK5 under VM/CE and give
Claude instructions for what tasks to perform in the environment to do some
end-to-end testing.  Only this time we don't have to invent the file
transfer because IND$FILE is already a thing.

Adam

On Tue, Aug 11, 2026 at 5:25 PM Cameron Tyre via COFF <coff at tuhs.org> wrote:

> Purely from personal observation, I think there is a degree of peer
> influence as to whether one accepts or rejects "AI", assuming you have a
> choice, i.e. dependant on whether your boss tells you to use it or not. I
> know, in some jobs now, there is no choice, unless you count qutting.
>
> I was helping a family member set up a website for her business. Aside
> from the obvious necessities like name of the business and service offered,
> I asked, "Do you have a spiel you want to put, on an 'About me' or 'About
> us' page on your website?". I pictured perhaps a couple of modest
> paragraphs revealing relevant experience.
>
> I was handed a pile of printed sheets of paper containing many, many
> paragraphs of what soon revealed itself to be repetitive and mostly generic
> phrases that could have applied to many businesses. I pointed out that
> there seemed to be repetition and asked if that was a mistake, i.e. had a
> paragraph or four been copied and pasted by accident.
>
> I was told, "Ooh no, I just told ChatGPT all my hopes and dreams and it
> wrote it all for me!". That was my first indirect experience of "AI" and it
> made me see that many people use "AI" because they are short on time, don't
> know how to do the thing themselves, can't be bothered doing it themselves
> or similar reasons. When I remonstrated with my family member and said, "A
> shorter piece, actually written by you, would be more powerful.", I was
> told, "But ChatGPT has done it for me!".
>
> My wife and many of her colleagues use "AI" at work because they are given
> far too much to do by the KPI obsessed management team and it probably
> saves one or two of them from having nervous breakdowns before school is
> out for summer. My wife, in particular, uses it to create visual aids such
> as pie charts and graphs which, by her own admission, might take her a
> whole day to create on her own.
>
> I think "AI" can be useful in certain specific circumstances but I have
> seen some awful "AI slop" videos on YouTube which get many things wrong and
> the "creator" of the video either doesn't know any better or doesn't care
> or both. My main concern is that people might, over a generation or two,
> get so lazy and uneducated that they will believe anything "AI" tells or
> shows them, meanwhile "AI" will start scraping the output of other "AI" and
> the output will get less accurate and less realistic through each iteration
> until we end up getting videos on YouTube about "Easter Island was the last
> holdout of the Nazis", or, "NICAP had proof that the Moon landings were
> faked - 'lost' footage revealed after 60 years!".
>
> I bought a "Python for Dummies" book a couple weeks ago, mostly because I
> have fond memories of earlier books in the Dummies series. I was astounded
> that a large chunk of the first chapter was about how to set up the VS Code
> IDE and tie it in with a subscription to an "AI" account so that you can
> get the "AI" to write code for you. I felt that kind of negated the point
> of the book. I aim on reading the whole book to the end but how many people
> will try and find the shortest route to making "AI" write code for them?
>
> I was recently reading about the decommissioned Bradwell nuclear power
> station in England, UK. The article I read stated, "In 1969, a new
> Honeywell 316 was installed as the primary reactor temperature-monitoring
> computer; this was in continuous use until summer 2000, when the internal
> 160 kB disk failed. Two PDP-11/70s, which had previously been secondary
> monitors, were moved to primary."**  My word of caution, related to that
> story, a question of caution really, is, "Would mankind let 'AI' write the
> software for a nuclear reactor temperature-monitoring computer?"
>
> Best regards,
>
> Cameron
>
> ** A super-interesting rabbit hole, in case anyone wants to take a trip
> down it, specifically the Honeywell 316.
>
>
>
>
> Sent from Proton Mail for Android.
>
>
>