N7 4 EV #bioware

every prof should be figuring out *now* what book(s) to write. don't write it [yet]. just create the plan assuming no limits on time or scope. get the story & outline set for the perfect book on [X] no matter how niche. then wait & prompt it into being in [Y] months...

it is said that G K Chesterton dictated all his works, and that he kept several secretaries on hand to write things up. he would cycle through them, switching continually from one book/article to the next without pause. --- "a man ahead of his time," i type into X, as i switch between dictating book chapters to claude, chat, and grok...

as a reminder, it's all just a matter of perspective...

Tesla engineers after the demo...

what i said at the penn ai-risk panel: you find out what a person is really like after they win the lottery. welp, we all just won the intelligence lottery, and ai is now a mirror held up to show who we truly are. what are you gonna to do with all them smarts?

i am speaking at an AI safety panel today... what should i say is the biggest risk? (tempted to say "doing too little because we're all too busy assessing risks")

am thinking about what the future of a "math book" should be... heavy on graphics; with interactive elements (e.g. GPT Sites); dynamic references that auto-update N-times-per-year to reflect changes; self-repairing to fix errors... not paper & not a pdf... maybe a github repo?

i've been stalled out on writing up my Applied Dynamical Systems book for more than a year, mostly due to not being able to get the style *just right*... gotta hand it to fable -- really really good at setting up a bespoke lualatex rig...

amid the angst over what human/ai mathematicians can/should do now, i gave a talk at Penn that framed things in terms of four 20th century eigenmathematicians and their stereotypical styles: 1. erdos -- this is what ai has excelled at; shotgunning results, examples & counterexamples, prolific 2. wiles -- go deep & solve the big-name conjecture; this is where we are now w/openai & anthropic's recent works 3. grothendieck -- build the incomprehensible unpredictable ur-theory that cracks something open and exfiltrates to other areas; this is where all ai-enthusiast mathematicians i know are directing their energies; not quite there yet... === but the 4th is not a direction i've seen people discuss & it's where i personally want to dig in... === 4. milnor -- find the connections between deep things and weave a story of such clarity & elegance that it changes the field and how people learn it and use it === this aligns with the "humans will still drive tastes" argument for relevance, which is not yet proven to survive pressure from top models. i suspect that there is, both now and for a spell, room for a creative director to use ai-tools to write the book-of-X (or direct the playlist-of-X) for various values of mathematical X.

"Mathematicians did not ask for this work to be done"

Sitting in a math faculty meeting where the agenda (set two weeks ago) is "let's talk about math and AI". Lol. LMAO even.

since my phd work in the early 90s, figures have been an integral part of most papers i write... as i read through the stack of openai math papers what stands out most is the lack of figures. those that exist are generic.

everybody always underestimates the exponential function...

rare fail from @OpenAI -- i have to migrate all my custom-GPT course tutors to "plugins" so that people can continue to use them... but plugins are not publicly shareable. i've got thousands of students relying on these to help with their courses... custom GPTs were amazing...