Carl T. Bergstrom

@carlbergstrom.com

Professor, UW Biology / Santa Fe Institute I study how information flows in biology, science, and society. Book: *Calling Bullshit*, http://tinyurl.com/fdcuvd7b LLM course: https://thebullshitmachines.com Corvids: https://tinyurl.com/mr2n5ymk he/him

LLMs have reached the point that they can be useful as proof assistants. Increasingly, they appear capable even of generating entire proofs that are correct. But in my experience, their best proofs are just correct. They aren't explanatory the way a good human proof is. Here's Cris Moore:

All this puzzling is very similar to what we do in mathematics. I think of a puzzle—especially a puzzle with a beautiful logical path—as an elegant proof of its own solution. It may be presented in cryptic form, or as a series of subtle signposts, but it provides just enough information to let the solver find, or build, its solution.
To continue the analogy, the best proofs are not just a series of logical steps arriving at a conclusion. Such a proof might establish that the conclusion is true, but that is all it does. It provides mere knowledge, and knowledge only of that one thing.
In contrast, a good proof gives us a sense of why something is true. It helps us learn, not just about that one problem, but about the larger web of mathematics: what analogies it has with other problems and other truths, how far these analogies can be stretched and where they break down. It teaches us something new about how truth works. It gives us insight and understanding.

Pro-wrestling executive Linda McMahon issued a letter today (in her role as US Secretary of Education) demanding that universities justify their existence and explain policies to the Trump administration by the end of 2026. Not sure what the endgame is, but it sure feels like a trap.

U.S. Secretary of Education Linda McMahon Issues National Call to Action to University Presidents and Governing Boards

Today, U.S. Secretary of Education Linda McMahon issued a national call to action to university presidents and governing boards, urging them to reaffirm higher education’s foundational commitments to ...

ed.gov

I caught James McMurtry at the Tractor a few nights back. I've been a fan ever since 1989 when the college paper where I worked got a copy of his debut record. He came by his storytelling ability honestly; his dad was Larry McMurty. Here's one of the exceptional tracks from that first record.

Crazy Wind by James McMurtry

Listen now on your favorite streaming service. Powered by Songlink/Odesli, an on-demand, customizable smart link service to help you share songs, albums, podcasts and more.

song.link

Nature did a short piece on our preprint about how LLMs will affect the practice of science; my more detailed thread is below.

Scientists using LLMs will ‘do more, less well’, modelling study predicts

Research suggests that the combination of incentives to publish and the use of large language models will lead to more papers, but they will be less refined.

nature.com

Carl T. Bergstrom@carlbergstrom.com · 2w ago

1. We—@eduede.bsky.social, @mjcrockett.bsky.social, Kevin Gross, and I—have a new preprint on the arXiv today, based on ideas that emerged during an @sfiscience.bsky.social workshop in November 2024: The unintended consequences of large language models as a labor-augmenting technology in science.

Listening, yet again, to Chris Watson's El Tren Fantasma, one of the most wonderful soundscape recordings I've ever heard. For me, it's a travel staple. I can't explain exactly why, but I love listening to it while flying. Perhaps wishful thinking.

El Tren Fantasma by Chris Watson

Listen now on your favorite streaming service. Powered by Songlink/Odesli, an on-demand, customizable smart link service to help you share songs, albums, podcasts and more.

album.link

Nature did a short piece on our preprint about how LLMs will affect the practice of science; my more detailed thread is below.

Scientists using LLMs will ‘do more, less well’, modelling study predicts

Research suggests that the combination of incentives to publish and the use of large language models will lead to more papers, but they will be less refined.

nature.com

Carl T. Bergstrom@carlbergstrom.com · 2w ago

1. We—@eduede.bsky.social, @mjcrockett.bsky.social, Kevin Gross, and I—have a new preprint on the arXiv today, based on ideas that emerged during an @sfiscience.bsky.social workshop in November 2024: The unintended consequences of large language models as a labor-augmenting technology in science.

To get a sense of how vacuous and ill-founded this claim is, compare: "If faculty and students seek to pursue the truth wherever it lies, a university cannot abjure its responsibility to advance social change." Which seems more plausible to you?