Adding a feature because ChatGPT incorrectly thinks it exists | Holovaty.com

@csalzman this is really interesting! I’m pretty skeptical of “AI” anything, but there’s something here about the way LLMs work that could be a helpful “how does it make sense to build this?” because your own dog food doesn’t cover everybody’s use cases.
Obviously it fails as a solution-provider when it makes up methods in well-known Java classes, but it seems like it was a good problem-provider in this case!

@csalzman obviously it might just be a gimmick, but it seems like this makes their service more useful for a bunch of people, and adding the ASCII thing was probably easier than what they had built for sheet music to begin with?

The author grapples with this a little, but: are you being forced to add a feature you didn’t want, or nudged to add a feature that you didn’t realize users would want?

@jwschultz right! It’s fascinating. In this case I think it’s a positive. It does make me fearful of a future where people demand we match what the AI is telling them is true. But I’m a curmudgeon