It was during the 17th month of the job search that I decided that my approach needed to change. All the good jobs were going to LLMs, I was sure of it. I decided to turn the tables and make my own LLM that could go get a job and then do the work for me. I dumped my digital life into Markdown: a few hundred tweets, a handful of college essays from my university-provided OneDrive, a protracted series of attempts to be funny in comments on high school acquaintances’ Instagram posts; a smattering of notes scattered across Evernote, OneNote, Google Keep, and Obsidian, a few dozen outbound emails, and about ten thousand words worth of arguments in Discord servers over whether Twenty One Pilots were ever cool. Then I got to work. Logically, to make an LLM whose work I could take credit for, I just had to make an LLM that was me.

My initial experiments were a little bit shaky: I found myself talking about punk influences in pop music to a befuddled interviewer over Zoom after the LLM picked up on music analysis as the most salient thing in its training data, and I followed along with the suggestions it gave based on the voice-to-text transcript it was getting until I didn’t quite know how to circle back to the product manager job I was hoping to snag. When LLMs talk through you, you never know quite what you’re about to say. Still, it hadn’t gone much worse than my last interview; at least I hadn’t brought up the JFK assassination. Chagrined, I turned back to my resume in the hopes of hammering it into the shape that would get me that next chance. Wait a second, I saw myself say. Let me try fixing it.

I watched, bemused, as me-the-LLM took over me-the-human’s terminal and started editing vigorously. In mere moments, the file was full of cliches and buzzwords; formulaic, mindless, barely-parsable phrases like “Fluent in Angular as well as rounded border radii” and “deep knowledge of the programming language Git as well as Github and Git++.” The thing is, it worked: as soon as I stuck it up on Linkedin, recruiters started contacting me en masse, enraptured by the number of boxes that my resume checked. Before long, we had more companies reaching out than we knew what to do with.

“Wait,” the next interviewer said as I typed in his questions and then read my responses off of my phone. “Am I interviewing you, or am I interviewing ChatGPT?” “You don’t understand,” I said. “I distilled an LLM out of Llama 4 using my own thoughts, ideas, and personhood. The LLM is doing the thinking, but the thoughts are all mine.” “You can’t take credit for that,” he said. “Sure I can,” I said, and to illustrate my point, I held up my phone. “Sure I can,” read the phone. “I can absolutely take credit for everything this dumbass is saying. He can’t even think without me.” It turned out that not all of the original training data for Llama 4 had been distilled away - mostly, making it read through my writing had just trained it not to be helpful, harmless, or kind. The interviewer didn’t seem to know whether that was a point for me or a point for him, but either way, I didn’t get the job.

My subsequent attempts to get myself employed as an LLM were rough - it turns out that they’re expected to be able to pass numerous genius-level benchmarks, despite the fact that they can also be convinced that the sky isn’t blue, and I failed to train myself on leaked answer keys thoroughly enough. Before long, I was in my 34th month of the job hunt, and the LLM was in its 17th. In retrospect, it was my mistake - in attempting to make a version of myself that was employable, I made a version of myself that was a little too close to what was already out there. I can only hope that one of us will distinguish ourselves from the unprocessed, undistinguished noise of LinkedIn posts and Internet arguments someday.