Last year I gave a presentation about Workflow in Sitecore that included a companion 3-part blog series. The final post involved developing with workflow, and when I was finished I felt a little underwhelmed. I was new to mutations and I felt like I could provide a better example.

I turned to ChapGPT, and I was initially quite pleased at the result. Where my example mutation only modified a workflow’s state, ChatGPT returned an snippet that could actually execute workflow actions. It was such a good example, I thought that this was something that was baked into the system that I just didn’t know about. To dig a little deeper, I went ahead & googled the name of the mutation, ExecuteWorkflowCommand, adding the keyword Sitecore

As is often the case when I’m googling something Sitecore-related, the top result was a blog post by Fishtank Consulting: Sitecore Workflow Automation with Management API, by Sohrab Saboori.

Comparing Sohrab’s script to the ChatGPT response, it’s clear that this was not just an intuitive title for a common use case. ChatGPT was stealing code from Sohrab and giving no credit. Worse: if I had been moving quickly and decided to sloppily pass of ChatGPT’s work as my own, I would be stealing code from Sohrab.

I don’t think I’m revealing a big secret when I say that LLMs are plagiarism machines. I also think it’s pretty obvious that you shouldn’t just copy paste from an LLM with no edits or followup and attempt to pass that work off as your own. Still, this came as a surprise to me.

It’s really tempting to use LLMs for a lot of stuff these days, and as I’m wrapping up this post I can feel the urge myself. I think my opener is a bit weak, the tone could be more professional, and I’d love to wrap this all up with a nice business takeaway - something like “this all goes to show that while AI can be a useful tool, something something pay attention don’t steal.”

But the line feels a bit more blurry these days.

A quick copy paste of the whole article saying “make this sound more professional” would be so easy to do, and might make this sound better. Would that be passing the LLMs work off as my own? If it is, it’s definitely something I’ve done numerous times. Maybe I did that already, would you even notice?

You’ll have to trust me that I didn’t. Rather than have use my robot editor, I’m going to stick a style I’m very comfortable with: vaguely gesturing towards some larger concepts without answering any of my own questions.