Some Things I Just Think

“Some things I say. Some things I write down in my journal. And some things I just think.”

Many years ago, a colleague said those words to me semi-facetiously as we were talking about dealing with a difficult coworker. I’ve been thinking lately about how that applies to LLMs, especially since the launch of GPT-5 a couple weeks ago and its fast-follow update to make it “nicer.”.

I follow some basic rules when I’m using ChatGPT. I’ve turned off its ability to train on our conversations (and just hope that’s actually true). I’ve turned off its ability to reference saved memories (same). And I follow that semi-joking practice from a colleague: some things I just don’t put into writing for the chatbot to respond to.

But I’m coming to this pseudo-relationship from a place of privilege. I’m using ChatGPT as a professional tool. I’m putting data in and getting data out. I’m not asking it for insights, recommendations, or to be a sounding board on big, personal questions, because I have people in my life who I can have genuine conversations with when that is what I need.

Those conversations may be weird and awkward and harder than they would be with a probabilistic token predictor, but they also then have real give and take.

But not everyone has that. 

I wanted to write the first draft of this post to say “have a professional relationship with your chat bot of choice.” And I still think that’s worth saying, but I also realize it’s advice that is somewhat glib. So I think really, my advice comes back to my colleague’s from many years ago. The same way you would with a person, when you’re using an LLM, think about what you say, and what could be done with what you’ve said.