Two stories about ChatGPT

Two days ago I tweeted :

I confess. I’ve succumbed and subscribed to ChatGPT Plus.

Eufrasio asked me why, so I replied

It’s great for translating code from one programming language to another. For example, I had a version of Gbloink! running in the browser, I wrote in CoffeeScript 10 yrs ago. There’s no future in CoffeeScript, but just translated it to Python, with GPT doing about 90% of the work

This is true. I’m hoping that I can release a new Python based version of Gbloink! this year. The Python is running in the browser thanks to PyScript But I should also be able to use it for a desktop app. I may even look into Kivy

As an aside if you haven’t seen them, you might be interested to see that I’m actually having a lot of success now with the original (1997 era!) Gbloink! running with modern DAW software :

Anyway, on the main topic of ChatGPT, Abraham asked

How’s it looking so far? Worth every penny?

And I answered :

Currently (since payment) it’s given me some code to add a) auto-complete to the “move card” field in Cardigan Bay. And b) to check broken links in a page.

I haven’t integrated them in yet, but I hope to in the next couple of days. If they work it’s a definite productivity boost.

Which brings us to the second story, as I tweeted today :

Damn! OK. So #chatgpt just bullshitted me (or at least helped reinforce rather than correct my own stupidity)

I wanted to write a script to check broken links out of my wiki. And I got Chat to help me write it in Clojure.

Of course, I was thinking it would be nice to have this running in the browser. A handy component at the bottom of pages that checked all the external links on them to tell me if they were still valid.

So after some back and forth, it seems like me and chat got the code working.

And then I hit the CORS problem …

Of course you can’t run this in the browser. And when asked directly, #chatgpt KNOWS you can’t.

But it’s happily accompanied me along the path of writing a nice interactive Reagent based widget for doing all this work in the browser.

Now I thought I was smart enough to use #chatgpt and NOT be caught out by its bullshit.

And I now realise that’s bollocks.

Yes, I can totally be fooled by it.

I hope I’m wiser now.

Which I think is a good summary of where I am.

Yes, ChatGPT is an incredible productivity boost. And, it’s even good for “rubber ducking”, ie. having “someone” to talk a problem through with. I’m not regretting subscribing. BUT … you ALWAYS have to be alert that it can confidently assert things that aren’t true or don’t make sense.

Leave a comment