Some Brief Thoughts on AI Writing

Depending on the context, I don’t mind using AI to write things. A lot of my classmates would do well to run their emails through chat first, just to get some politeness. But I really, really dislike reading AI generated/assisted content that attempts to have some real intellectual/emotional meaning to it. Two examples today:

1. A LinkedIn post from the mother of someone at my school, with a supposed lesson along the lines of “you don’t realise how special the ordinary moments are until they end”. While LinkedIn is obviously a slopfest, and I admit to using LLMs for some posts that don’t have a call to action or deeper meaning, pseudo-intellectual messages like that give me a sour taste when they come from a bot.
2. This speech on AI agents and competition policy by Andrew Leigh, an Australian MP. About halfway through reading it, some parts of my brain were telling me it could have been LLM-written. I wasn’t sure, because it was a speech and thus likely to have the sort of “mic drop every sentence” language that LLMs produce because of the form. But alas, it was indeed mostly written with AI, at least according to Pangram, which I trust.

I love using language models for research and notemaking. I also love them, sometimes, for the final output. The key issue with using an LLM as the final step is reader trust: how can I trust anything in a piece of content if I don’t know whether the author or speaker even has the capability to rephrase the LLM gobbledygook in their own words? Leigh could have asked ChatGPT to write that speech from a 50-word prompt, ten minutes before he got up on stage. It’s also possible he spent hours writing the speech and just ran it through a chatbot at the very end to make it sound nicer. But there is no way for the reader to know which it is, so I can only assume the worst.