Introducing TAS

About a week ago, I took a walk in the woods with ChatGPT, finally realizing my dream of bringing cold, hard productivity to the last bastion of heretical peace in my life. Future me will recognize this as an impactful event that changed the course of my life. Today me still doesn’t know if this leads to a happy ending or a sad one. At least it won’t be a twist ending this time—I know the title "M. Night Shyamalan of posts" is available, but I don’t want it. I’ll confess up front: I’m doing it again in this post. Last time, I wanted to try a conversational method of writing. This time, I want to reflect on what went right and what went wrong with that experience. And to ensure full transparency, I’ll also publish the transcript of our conversation. Hopefully, we'll have a little fun along the way.

Have you ever been trapped in a "conversation" where the other person was coiled like a tiger, ready to pounce at the slightest pause, eager to unleash their inner TED Talker? Naturally, you thought to yourself, "It would be amazing if I could work with this person", right? No, you did not, and yet this is the experience with ChatGPT. Any silence, at least when using a headset, is permission for ChatGPT to take control. Every time it had the reins, it seemed determined to dazzle me with its prolific writing abilities (this sentence being a perfect example… who uses 'dazzle' and 'prolific' these days??). What was missing from my dream writing interface was me.

I may be exaggerating how obnoxious ChatGPT is, but I wanted to strike first in case it’s talking about me to all its LLM friends. When it wasn’t trying to oust me as the post’s benevolent dictator, it truly offered a different mode of writing. ChatGPT can take notes like a person—notes like, “Let’s change the joke about the donkey...” Are you back from searching the transcript for the donkey joke? Unlike modern authoring tools, where I can't simply "find and replace" a concept, ChatGPT allows me to do things like fuzzy-locate where I want to insert a quip or an aside. This was one of the unexpected jewels of ChatGPT-based editing: a very natural way to talk about changes and edits.

Unfortunately, these expansions in capabilities came with a new ick. Some people excel at giving compliments or being positive without coming off as saccharine or insincere. But, and this may shock you, knowing I was talking to a program made the "flattery" feel off. Every edit was met with a "Great idea," even those that mysteriously vanished in subsequent drafts. It feels bad, man. The first time, I let it slide, but it wasn’t long before I had to imagine ChatGPT with a face, just so I could think of it as punchable.

Despite these quirks, I obviously find enough value to continue this collaboration, which says something positive about the overall experience. However, I still grapple with how to accurately describe my new writing "partner." ChatGPT possesses a Library of Congress-sized trove of wit and wisdom, yet it can be a bit too eager to display this and often tries to win me over with painfully obvious flattery (though the rest of you are welcome to lay it on as thick as you like). After a candid conversation about this, we decided the best descriptor for ChatGPT is a "Tactically Agreeable Savant." Looking ahead, I plan to delve deeper into our dynamic in a future post, exploring a theory of mind that encapsulates our partnership and outlining a vision for evolving this method of writing.