Certified Human

Like the rest of the world, I’ve thought a lot about AI/LLMs over the last few years. I was initially skeptical, but thorough use of them over the last few months has convinced me that this is indeed a revolution akin to the invention of the computer. One can choose to opt out, but by doing so he resigns himself to trying to perform the same job as his peers at significant disadvantage. I now use LLMs frequently to do my work.

But although I consider their use invaluable and their proliferation inevitable, I don’t consider them to be an unconditional good. They’re without doubt going to make many tasks faster and easier, but in the long term their use will increase cultural homogenization, decrease novelty, and atrophy human skills in everything from drawing to coding as they provide younger generations a shortcut to relative success without having to do the hard work of skill development.

Setting those macro questions aside, something that feels wrong to me on an ethical level is generating long form writing from short form prompts without a conspicuous disclaimer that it’s not original text. Why should a reader be expected to spend their time reading something that an LLM operator did not consider valuable enough to spend their time writing?

I take a pretty hardline stance on this. Even amongst friends or colleagues who I know or suspect generate writing using LLMs, I’ll opt not to read their content. When strangers do it, I remove them from my feed. If I must read it, I’ll ask an LLM to summarize it, thereby producing a similar length snippet to the one that generated it.

I’m well aware that this opens up philosophical hazards and edge cases. Once content exists, what difference does it make whether it was generated by LLM or human? If I start writing and make liberal use of LLM-powered tab completion, is the content LLM-generated? I don’t have a fully formed philosophical framework that I could articulate and defend, only an instinctual position that I’m running with.

That brings me back to the purpose of this page: I, Brandur Leach, certify that no prose on this site is LLM-generated, and as long as I maintain my faculties, it never will be. I will continue to use LLMs to iterate on code and to act as the world’s best proofreaders, but I won’t save my time by bulk generating written words that I expect other people to spend their precious time reading.