Earlier this month I wrote about why it seems AI likes archivists.
I sometimes feel bad about how old some books on my to-read stack are, particularly relative to their recency when I first bought them. But, of course, more books keep coming out so what can you do! I just got around to reading Zadie Smith’s On Beauty and I’m actually glad that I waited to read this until I, and many of my friends, are also academics in our 50s.
As this article acknowledges, from a data science side the processes used for watermarking AI generated text are interesting and it’s sort of cool that it works. But the sentiment of the title, Anthropic’s Text Watermarking Proves AI Companies Do Not Care at All About Writing, gets at an important point that you need to know what a system was designed to do to understand if and how to use it. If word choice is treated as something arbitrary that can be manipulated without changing the quality of the text, that tells you something about the types of texts you can reasonably generate.
That article is nicely complemented by the following: In Praise of “Wordy, Old Prose.” (Or, F*ck Your AI Reading Buddy)
The complete collection of Slings & Arrows is available on YouTube.
Before there was AI, or even spell checking built into every application and app we use, you could spend $50 (in 1983 dollars, the equivalent of a bit over $150 today) to buy the Random House spell checking software – number of words available depending on your hard drive and memory capacity:









