Thoughts on Gruber's criticisms
Yesterday John Gruber posted a take on Anthropic's text watermarking feature that attracted a lot of attention by making it clear he doesn't understand how LLMs work. Today he posted a reply to all that feedback, adding more arrogance to his ignorance.
"I must disagree that it can possibly be seen as a net good for LLMs to produce worse prose." he says.
Complaining that the function that randomizes your output is not objectively perfect because it's injecting some non-random noise into the random noise generator is a strange position to take.
His views on writing and writers in response to Dan Moren are even more uninformed, but this is probably not the place to unpack post-modern literary theory. But trying to understand the automatic sentence machine with a normative Modernist worldview is like bringing a philosophy degree to a knife fight. Pointless in more than one way.
And while I have no ethical problem with machine generated text in general (aside from the obvious burning the planet down bit), the point of writing, for me at least, is only partially the output. Writing is a process that forces clarity onto the thoughts of the writer. Perhaps John should do some actual writing on this topic and see if that helps him understand it a bit better.