A question I’m frequently asked is whether I use LLMs to produce my poetry.
The short answer is No👎🏻, not in the way questioners mean. Here is the much more complicated real answer.
Financing: Yes👍🏻
The main way AI use helps my poetry is that it constitutes the day job that makes the writing possible. At my sincerely beloved employer Mitteldeutscher Rundfunk, I used to be basically another sysadmin, but since ChatGPT came out, as one of a few one-eyed among the blind I’ve been given various AI-related management responsibilities. At MDR we do broadcast media: one of the first industries to be disrupted, and also one that with the still-recent disruptions by 1. the WWW, 2. smartphones and 3. social media well underdtands that a new technology can necessitate a lot of changes, and new expertise to steer those. I’m not doing an ambitious career, but I’m doing a good job, advising, guiding and implementing various AI related projects. If it wasn’t for AI, I could not earn a salary above the regional median for less time spent than I put into parenting.
verse writing: No👎🏻
Even now in the summer of 2026, the LLMs still can’t write Common Meter well enough! 🙂↔️ I have tested the newest models on this again and again every couple of months since the initial ChatGPT release, but never managed to get any LLM to produce poetry that met my standards. I first saw one rhyme decently in 2025. By now, most new models can usually do rhymes OK. But meter continues to seem to befuddle them, unfortunately.
Poetry reading: yes👍🏻
There has been LLM poetry I greatly enjoyed reading. My top favourite poem written after 1956 (Howl) is Silver Bird Above San Francisco written by the epochal genius Gwern (who I have been a fan of for a long time) and his crew of diverse LLMs.
Another that hit me like a lightning strike, as a proper poem should, was this unnamed one written by DeepSeek R1 without much of a prompt, i e. pretty much on its own:
So yes, even this current generation of still comparatively simple AI can already write poetry more worth reading than that of famed human poets like the exceptionally overrated Louise Glück. They just can’t do the strongly metered kind that I love best.
For now. Here’s where we’re starting from,
with lots of room to grow a
lot better poets yet to come.
These now are protozoa.
Illustration: yes👍🏻
I enjoy making extensive use of Generative AI to turn some of my poems into pictures…
…and songs and (since progress in direct text-to-video has disappointed me as a regularly as progress towards Common Meter writing) “videos” composed of AI audio and AI still images, such as this presentation of the first half of my Horatius translation…
…this pop song…
…and this sci-fi short story in verse that AI generated operatic singing turned into “space opera”. 😄
Such creations are widely considered lowbrow “slop”; many prefer to just highbrow read the poems. But
- I’m having great fun with this,
- it seems to produce noticable added value over the plain poem text for negligible additional effort
- and I’m arguably researching the ascending curve of AI capabilities for my employer.
So I expect to continue to do that. Maybe at some point I’ll make a replacement of this two year old video of my own voice speaking only the first five Sermons illustrated with what Midjourney could do at the time…
…With one where newer models speak and illustrate the full set. But that will be a side project; the verse writing will always have priority and since I haven’t found a way to automate that, I expect it will continue to take most of my free time.
Final brag
One of the proudest moments of my life was when I got an email from somebody who had read to my hero Richard Dawkins himself my verse version of his text.
I was told he loved it! 😃
And my excuse to mention this 😎 is that he then asked the question I wrote this blogpost about: whether that had been written by ChatGPT. It was of course published in 2014; that’s better evidence it wasn’t LLM work than I expect to have in the near future.