I’ve been reading Katherine Randell’s excellent Super-Infinite: The Transformations of John Donne. (Check out her interview by Tyler Cowen, it is delightful.)
Early in the book she quotes his Satire II, in which Donne criticizes a particular plagiarist and lawyer.
But he is worst, who (beggarly) doth chaw [i.e., chew]
Others’ wits’ fruits, and in his ravenous maw
Rankly digested, doth those things outspew,
As his own things. And they are his own, 'tis true,
For if one eat my meat, though it be known
The meat was mine, the excrement’s his own.
Donne is referring to the remixing of ideas by a human intelligence. But I’m steeped in AI news these days, so this jumped out as an obvious (and relatively accurate, I think) metaphor for large language models (LLMs). A LLM ingests massive amounts of others’ words into its “ravenous maw,” digest them, and then “outspew[s]” the result as its own.
Donne would seem to be no fan of the end product - he calls it shit! Yet he believed the derivative creator still owned the results. No dinner host would claim to own the waste generated by their dinner guests, after all.
Digestion isn’t a perfect metaphor for LLMs. Digestion breaks down the connections between various food components. In contrast, LLMs identify connections between tokens. But both fundamentally transform the original works, for better - or, if you’re Donne, for worse.
A few outputs from this week:
🎤 - Debating The Conservative Approach To Generative A.I. — I talked with Emily Jashinsky on the Federalist Radio Hour about AI, last week's Senate hearing with OpenAI's Sam Altman, and what the U.S. can do -- and shouldn't do -- to maintain its lead in developing and deploying AI technology. (link)
📰 - Don’t Pause AI Research; Pause AI Regulation — Chris Koopman and I had an op ed in USAToday on why Congress should take a breather before plunging in to regulate AI. (link; it’s paywalled, sorry)
🎤 - The Future of AI Regulation — Adam Thierer and I joined the Regulatory Transparency Project podcast to discuss an “FDA for Algorithms,” AI licensing, transparency mandates, and impact assessments/audits. (link)
Thank you for reading, and I hope you have a meaningful Memorial Day weekend.