Today’s live stream was a little different: I demonstrated using the Aiomatic WordPress plugin, ChatGPT, and Azure Open AI to write blog posts. I showed the kinds of content it writes, the kinds of blog post content it doesn’t include, and taught you how to identify meaningless word salad blog posts written by people who don’t actually know what they’re doing.
I finish up by talking about the kinds of places where it actually is useful, and how I use it to wireframe out new work.
5 Comments. Leave new
A Huel shirt!! Awesome. I used to replace meals with that and would carry it in a bear bag when I would go hiking just in case. I got kind of bored at just drinking it, so, I experimented and found that waffles where you replace the waffle mix with Huel is a fun way to eat it.
Hahaha, yeah. I’ve been using it as a time-saving meal replacement for about a year and really like it. I feel like there are only two kinds of meals in life: the ones I want to remember (either due to the meal, place, time, companions, etc), and the ones where I’m just checking off my body’s needs.
For the latter, I just wanna eat quickly and be done with it, hahaha, and Huel is perfect for that.
I have been looking for a practical use for this technology for a while. I experimented with having it boiler plate code, but the quality of code it has generated has been so atrocious it was far more time to review and edit what it produced than it was just to write it. Getting code that doesn’t compile at all wasn’t uncommon in both PowerShell and TSQL. In PowerShell it had a tendency to use obscure aliases, highly version specific .net code, generally format the code so strangely that it was incredibly confusing to figure out what it is even supposed to do.
In the last few weeks I have been writing a SQL code style guide. I found that it was very good at producing polished sounding style guides that said almost nothing as your examples in the video were experiencing. When I prompted it to include more specific information – telling it that aliases have to be implemented with the AS keyword, commas go at beginning of line, etc, the content that it produced became more and more unpredictable, sometimes only reformatting what I asked it to write into bullet points. Whenever it provided examples, the examples ended up totally inconsistent and sometimes would provide a code example of a specification without implementing what I told it to.
The second biggest challenge was not knowing what I wanted in it – the biggest challenge was getting content out of it that managed to say anything at all without it turning into talking absolute garbage. I did end up getting a document format out of it that came from one of the better, but very incomplete versions of the document, however the amount of effort that it took to get anything useful was just insane.
I definitely see your point about using it as a wireframe. I am curious if it is suitable at all for trying to produce a document with specific requirements, (as opposed to writing something with a more open-ended goal) I certainly could have produced what chat GPT did for me far faster and then not need to go through to edit and reformat the document.
One already-practical use is to give it code, and ask it what problems the code might produce, or check it for bugs. It seems to be better at taking specific things and producing general advice – as opposed to the other way around.
Ohhhh I like that idea. I have a lot of garbage code from report writers and occasionally weird ORM queries that I have spent half a day or more on trying to figure out WTH. Am pretty excited to try that!