Machine Transparency
We have been writing this blog for ten years now, and I just feel so much happier about it now that I have the assistance of Large Language Models (LLMs). I love writing and deep thinking, but I’m not good at schedules, or deadlines, or finishing things. LLMs make the more difficult aspects accessible to me.
So, I figured it was time to write a transparency statement about how I use LLMs on the American Cyborg website:
The Excelsior! blog is the primary location of my LLM use; the rest of the site is human-made art and text.
I have been using OpenAI’s models since 2022 but am currently in the process of transitioning to Anthropic. In the near future I hope to have skill enough to use Hugging Face as well.
All our team members are assigned a bird during orientation; ChatGPT selected Magpie. Any blogpost that was written with the assistance of LLMs is tagged #magpie.
For all blogposts: the first draft is human-written, the middle drafts may or may not be machine-assisted, the final draft is human-written. When a blogpost is still in drafts, the author is listed as American Cyborg. When it’s finalized, a specific bird is listed as author.
I like to have both a machine-made and human-made image for each blogpost. Using LLMs to generate images is one of my favorite things, I find it absolutely magical. And sourcing images for our blogposts has always been a challenge — getting a custom artwork with lots of constraints and rapid turnaround isn’t fun or worth the money for me or the artist. I like using my own photography, but I have to set up the shoots in annual batches.
I do treat the entire blog archive as a living document. I preserve all the drafts of all the posts, and I never misrepresent our past writing. But I haven’t fully synthesized past implementations of our blogs, notes, meetings, etc, so things are getting added and rephrased as I go.
