Oh, Lord, Don't Let Me Be Misunderstood
Sometimes the best thing an LLM can do for you is misunderstand you
Both Eric Burdon and Elvis Costello knew what really gets our energy up1
I’m not teaching writing (or anything else at the moment). I’m working a management job on a university campus.2 I realized this morning that one thing I do in my “job job” might be useful for some people. It’s useful to me at any rate.
When I am writing a piece explaining a process, policy, procedure, or project, I’ll write the bullet points of that explanation out myself. I almost always do this manually, I haven’t found AI helpful at this “sketching” stage.
I then take the bullet points and ask AI to write the email, policy, or whatever.
Ah, you say, this is how everybody uses this. Not quite.
I then look at the AI text and use it to go back and refine my bullet points. The LLM output for me turns out to be a great way to surface ways my writing can be misunderstood as people fill in the gaps. It’s honestly the errors of the LLM I am interested in most. Sometimes I’ll reverse engineer the text back into corrected bullet points, and sometimes I’ll just re-edit the original bullet points.
Then I’ll write the thing myself.
It sounds like a bizarre way to do things, but there’s something about seeing my stuff misunderstood on the page that gets the dopamine and adrenaline up, and helps me move forward.
I know sometimes when I post things like this people (not necessarily here, but in places I share these posts) say ha, ha — it seems like a lot of effort to write something. And it is, I suppose. I just generally feel that writing anything where the dopamine isn’t there is a lot of effort for me. I’m a famously quick writer when the synapses are firing and a slow one when I’m bored with the thing. Having an LLM misunderstand my work gives me something to react to that pricks me out of my stupor. Your mileage may vary.
I need to learn more about the PAIRR framework, but I think there is something similar there.
I am 90% sure it was my high school friend Chuck Gabriel who introduced me to this song, and I find it really gratifying that he follows this blog, and find it lovely that he’s now teaching future generations of musicians at UMass Lowell.
I want to get back to teaching and research, I think, but I’ve also found that I’m a good manager and am really enjoying managing a team of talented academic and classroom technology professionals for the moment. It’s been a weird few years.
Anyway, I’m mostly writing this footnote because I think a lot of people are confused that when they reach out to me I can only schedule early morning or weekend meetings, or do keynotes on vacation. I have a job! I will reply to your email (maybe) it just is going to take some time!

Thanks for describing this dynamic so precisely! I often feel this. It's one of my wishes for students that they get to experience it, but in my first-year composition classes I think a lot of students need to build up confidence and history of success in getting to greater clarity through writing first.
Thanks for thinking of PAIRR! We definitely want to figure out how best to support students to get there...the approach we take is to require students to chat back and to encourage them to disagree with the feedback they get and give them sample phrases for pushing back.
I talk about it in this slide deck if you're curious: https://link.annarmills.com/feedback
One further approach I've been toying with is inviting contradictory feedback. So far I think this might be just too much to process for students... but maybe I'm wrong. I made a chatbot that calls for three mutually incompatible responses to every query (in the slide deck).
In PAIRR, we hold off on inviting students to use AI to generate text that fulfills the assignment because we want to protect that development of confidence and agency around writing...
What can I say? A shoutout from you is a great way to start the day! I recommend it to anyone!