Kameron Decker Harris

The many guises of procrastination

what Piali said! Looking to recruit a new assistant professor to our fabulous neuroscience group and department at Brandeis. Please circulate far and wide and alert anyone you know who is on the job market!
#neuroscience

Niceeeeee. I may be getting out for a first foray of the season this weekend!

No kings allowed!

Allen Institute - Shanahan Foundation Fellowship at the Interface of Data and Neuroscience 2025-2026 Thank you for applying for the Shanahan Foundation Fellowship at the Interface of Data and Neuroscience. Applications for the Fall 2026 cohort will close at 11:59pm PT on December 1st, 2025. Important...

Theory/Computational post-doc position at the Allen Institute. This is a unique opportunity to do theory in a community deeply embedded in data.

alleninstitute.submittable.com/submit/33511...

Hey Microsoft, before you transform humanity with artificial general intelligence, can you just get the search to work on Outlook?

Without sleep, we die. 🧵

Options within a machine translation software?

Silas sighting!

Here you go: www.jmlr.org/papers/v25/2...
You might also like our work where we showed similar effects (and more) in a related model: openreview.net/pdf?id=bvJ4U...

What's your policy on AI usage in your class? I'm doing some short form in-class writing assignments where I can be sure they're doing it with just pencil, paper, and brain

It's not about contextual modulation, but it makes the argument that the weights you find in standard deep networks for image classification are effectively low-rank and that the same goes for the feature embeddings

😂

Neat, I'm mostly aware of the work in earlier models for things like vision, including the so-called "rainbow networks" paper by Guth et al

There have been a couple papers showing that low rank weights are common in deep nets, although I'm not sure of work in that area for LLMs

It kinda makes sense if you believe that weights already are fairly low dimensional/rank, so that a lower-D perturbation could have a major impact on the network function

I get that. It was fun for me as a sci-fi epic and academic satire

Redwall was formative. Did you check out Anathem by Neal Stephenson?

Dystopia today

Massive immigration raid on Chicago apartment building leaves residents reeling: 'I feel defeated' The Department of Homeland Security said federal agents with Border Patrol, the FBI and the Bureau of Alcohol, Tobacco, Firearms and Explosives arrested 37 people in the raid.

"Armed federal agents in military fatigues ... pulling men, women and children from their apartments, some of them naked... Agents approached or entered nearly every apartment in the five-story building, and U.S. citizens were among those detained for hours."

www.wbez.org/immigration/...

Very cool! You'll be our neighbor to the north. I'd love to reconnect... perhaps over delicious food in your new city ;-)

All the best

Great thread here!

Jevons paradox - Wikipedia

Jevons paradox! en.wikipedia.org/wiki/Jevons_...

Andrew Ng: "AI is the new electricity!"

Cory Doctorow: "AI is the asbestos we are shoveling into the walls of our society and our descendants will be digging it out for generations."

¯\_(ツ)_/¯

(pluralistic.net/2025/09/27/e...)

Very cool, congrats!

I mean, this seems like the way. At some point journals themselves are no longer relevant for spreading info. They become a way to manage the peer review and as prestige stamps. As a small lab at a small uni, we cannot pay extortionate or even small APCs

This is madness

Cylindrocopturus munitus

The new view is great

Ah man, yeah but the caveat is there is no cooling in the office... because "efficiency." So I gotta keep the door open and a fan blowing in cold air when it's hot