Generative AI and homogenization

Among the legitimate worries about Large Language Models is that they will homogenize diverse voices.1 As more content is generated by LLMs, the generic style of LLM output will provide exemplars to people finding their own voices. So even people who write for themselves will learn to write like machines.

With the most recent update, my blog software added a built-in “AI assistant” which will fill in a paragraph of blog content in response to a prompt. It offers ten different options for tone, ranging from formal to provocative. Each option is accompanied by a suggestive emoji, like “🤨 Skeptical” and “😎 Confident.” They give different output, but all have the same ersatz algorithmic sheen.

Perhaps giving more examples of its output would be like telling you about a dream I had— but even my banal and uninteresting dreams are mine. The Assistant thinks that my book A Philosophy of Cover Songs was written by anyone but me, attributing it to authors I haven’t heard of. And it can’t fathom a fruit that resembles a pot of giraffe stew.

  1. In an article from the distant world of last April, Sigal Samuel summarizes the dangers.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.