Last week I attempted to use ChatGPT to write a data analysis program in MATLAB for me...

I had MANY features I wanted the program to have, but I started with the most basic and thought we would add features slowly... ChatGPT 'agreed' this was a reasonable approach when I asked for its opinion...

So I asked it to create a program that popped up a graphical user interface with a 'load file' button, when the user presses the button a file explorer window pops up allowing the user to select a '.csv' file to read, and then read it, and pop up a new window and display it as a spreadsheet.

To my surprise it did this with no trouble... the resulting program was FAR shorter than if I had written it myself (It has been years since I last wrote a program of similar complexity- and I write code in unfamiliar languages by looking up examples of code that does something similar to what I want and then adapting it to my purposes....)

It wrote ~40 lines of code with comments that I copied and ran in MATLAB- it worked....

I then pointed out that some of the files it would have to read were 100's of thousands of rows long with 28 columns - so memory management was a concern... I asked if it could modify the code so that it dynamically only read into memory the rows the user was viewing in the spreadsheet- I think it did that ... the program seemed to run a little faster.

Then I explained that columns 2 and 12 were date/time values in the form of "yyyy-mm-dd hh:mm:ss.ssssss" I asked it to convert these columns to epoch time- it knew what I was asking and knew the commands to do it, but this is where it all started to go downhill... the code threw errors , as it tried to generate new code to fix the errors it started to 'forget' the features we had agreed upon earlier, so the new code lost those features.

I could tell ChatGPT was getting confused... It would also behave like it was afraid to contradict me - when discussing the format of the data in the spreadsheet I accidentally said "Rows 2 and 12 are formatted to be date/time values, correct" (I meant columns, not rows) - rows 2 and 12 clearly were NOT all date/time formats, but it agreed with me anyway... we talked about this, and it apologized, but never explained why it would agree to a clearly false statement.

The confusion increased and it just became impossible to proceed... so I started over from the beginning, hoping to avoid the same confusion... but after a certain degree of complexity we ended up in a confused loop... I would ask for a function to be added to the program and it would add it, but the program would throw errors... in fixing those errors it would remove that function, or a previous one it had added...

It was like it had the memory of a goldfish - forgetting what we had agreed upon in previous stages of the program... it was TOO agreeable- accepting things I said even if they were wrong...

So I was impressed, but it is NOT at the level needed for any moderately complicated programming work. I do think the NEXT generation will be far more impressive... and its just a matter of time before it can do all I need and far more. Perhaps it would do better if I asked it to write in PYTHON, a more commonly used language.

The social implications for all this will be devastating - AI will soon put many people out of work. What happens when vast swaths of educated and uneducated people alike lose their jobs to AI?