ChatGPT redux

the word chatgpt is spelled out in scrabble tiles

After my first experience with ChatGPT, I began to wonder, how far could I go with it’s programming capabilities?

My plans for a GUI based Image Importer were outlined in the previous post, so I won’t go into that. The descriptive shown last time was saved as a file and uploaded to ChatGPT.

The first run through was fairly uneventful. Per ChatGPT’s recommendations, we progressed through the process one item at a time.

  • The GUI was crafted. This was getting all the bit and pieces as shown in the graphic I previously presented.
  • Once that was done, we did the image read and thumbnail display process. That went fairly smoothly, with mostly being getting the images displayed properly in the thumbnails.
  • The next step was the assignment of “Descriptors”. Once a group of images was selected, the user would right-click and a dialog box popped up to either type in, or choose a descriptor from a list.
  • The next phase was the mockup of the renaming process, with a presentation of the new filenames in the right window.
  • And finally, the actual importation of the issues.

This all went fairly well. ChatGPT was building all this code in one file, so I tasked it with rearranging the file into sub routines for more effective management. It enthusiastically agreed it was a good thing, and proceeded to create a more logical project structure. I downloaded the zip files, expanded everything, and then the explosion occurred.

Running, or attempting to, run the file brought forth a plethora of issues. We worked through some of these.

It’s fairly well documented, that after long sessions with ChatGPT, it will start to hallucinate. We’ve reached that stage. In the last couple of days, working toward building the app, it’s been going effectively no-where. Regressions, code errors, etc, etc, etc.

My son, who is a programmer, mentioned that his organization has run into this frequently, and it has to be “managed”. So, we consolidated the session into a “specification” to start a new session with.

So, a new session was started, with the specification created in the previous session.

Good progress had been made with the tool. ChatGPT offered a free month of usage, so I accepted that. Previous to that point, I was getting about an hour a day of interaction before with would stop.

Being able to just work through without the daily limits was very productive. We got the tool to what would be known as a beta.

The CAE Carshow was the next photo outing, so I grabbed the X-T2 with the XF10-24 and went. There were some 125 raw/jpg pairs in the tranche, and truth be told, the tool worked well. Except….

There was a fairy significant bug where if you mouse out of the transfer window, it resets everything and starts to scan the source location again.😳

If you leave the mouse in the window, the transfer works just fine.

There had also been a regression in that the image thumbnail window doesn’t scroll down as the previews were loaded.

And then…….

I was this || close.

The routine was done, all the bugs swatted, very useful. And then I got greedy.

Asking for some small cosmetic tweaks, ChatGPT lost its mind. I wasn’t smart enough to realize what was going on until it was a complete mess.

So I told it to regress to the known good point, but it really wasn’t able to do that.

My son is a programmer, and he laughed at me. That’s what GIT is for, Dad. Learn how to use it…. 🙄

Sigh. So I put the project aside for a couple of days. Coming back to it, I told it to recreate the files from a place before everything went fluey. It’s clear that there is little saved history in these processes, and going back in a process is challenging.

But, we got to a good stable point, and then carefully moved forward, successfully completing the process. Nirvana. I’ve done some testing, and it seems to be robust. Hooray.

Here is what the GUI looks like:

I’ve created a page with the help file text: Image Import Tool.

At this point, the output format is fixed.

The plan is to take this and then create a more general purpose tool with user configurable output schemes, thumbnail sizes, etc, etc. A goal at this juncture is to make it OS independent. It’s Windows only now due to its usage of exiftool to write xmp sidecar files.

This has been a very interesting experiment thus far. While I possibly could have written the thing on my own, it would have taken many months rather than many days. It’s key that to do anything of significance, the process must be carefully managed. I’m a Project Manager all over again, but with ChatGPT being my only worker…..

On a separate topic, the photos page on the website is currently being expanded to include all the images currently in my media library. It will probably take a while before all that is done.

The plan, at this stage, is to go backwards in time from the CAE CarShow and “fill in the gaps anti-chronologically wise”. When new photo opportunities arise, they will be presented.

So, it will look very much like no rhyme or reason in the process. There has been a lot of photo activity just in 2025, so next time we’ll have a look see at the Jamestown-Scotland Ferry I’ve got a goodly amount of images of our rides on the ferry, so I’ll prolly wrap them all up in one post and accompanying page.