One of the greatest minds of our times has died. This is a huge blow to the fields of philosophy, morality, consciousness and intelligence. I adored his teachings even though sometimes it took me years to get them. His ideas will live on. Daniel Dennett

Daniel Dennett: How to compose a successful critical commentary Rest well, and thank you

There are several passages in Dennett’s books that I vividly recall where I was when I read them. He was perhaps my favorite living thinker. I never met him, but through his writing he felt like a friend and teacher. I’m very sad, but comforted by all the love for him in seeing.

Our journal will dedicate a special issue to the philosophy and legacy of Daniel Dennett, entitled Dennett's Dangerous Ideas. Details about the call for papers will be available soon. Watch this space!

Losing the smiling spirit of consciousness studies. Dan Dennett has passed, and with that, the consciousness studies has lost a source of inspiration. At a time when philosophy was passing through a narrow bottleneck of self-limitation, Dan had a real breadth of interest, while

Very excited about the work my student, John Hood, and I are presenting at #AISTATS tomorrow! It's a new form of tensor decomposition, called ALL0CORE ("allocated L0-constrained core"). The idea is simple (see fig), and it works really well. Read more: arxiv.org/pdf/2403.06153

New paper in FAccT 2024 with Aaron Schein, Yixin Wang Nikhil Garg! Many discrimination auditing and electoral tasks use ML to predict race/ethnicity – by discretizing continuous scores. Can the discretization process cause bias in labels and downstream tasks? Yes!

As an optimization problem, we formalize this problem of discretizing probabilities and its tradeoffs and find that a joint optimization approach can mitigate these distributional skews with negligible loss in individual prediction accuracy, while using *no additional data*.

New paper: How can you tell if a transformer has the right world model? We trained a transformer to predict directions for NYC taxi rides. The model was good. It could find shortest paths between new points But had it built a map of NYC? We reconstructed its map and found this:

Why is "token" used to mean "type" in the LLM lit? I'm used to it by now, but curious if there's a good reason.

In a tensor, the super-diagonal entries are those indexed by (i, i, ..., i). Is there a term for all entries of a tensor whose indices contain any number of repeats? e.g., in 4 modes, (i, i, k, j), (i, i, i, a), etc.?

How can the same LLM pass AP math and also claim 9.10 > 9.9? Our #ICML2024 paper: LLMs are wrong in ways that people can't predict. Benchmarks can be misleading because people decide how LLMs are used. Solution: measure LLM alignment with the "human generalization function" 🧵

Excited to announce the NeurIPS 2024 workshop on Behavioral Machine Learning! Are you working on AI methods that incorporate models of human behavior? Submit a paper by Sept 10! We invite submissions across CS fields and the behavioral sciences. More: behavioralml.org

I'm on the job market this fall. I study how the state and marginalized populations interact in American politics and policy, with a focus on policy implementation (state capacity) and community organizing and collective action (civic capacity).

Hey all! We're hiring this year at UChicago Poli Sci for an open rank position in quantitative methods! Happy to answer any questions.

You can read the accepted manuscript at tandfonline.com/10.1080/106186… and the arxiv pre-print at arxiv.org/abs/2408.14604 Minor differences between the two but the accepted version analyzes a larger dataset!

Am I the only one getting much more of this type of garbage on my X feed? My feed started noticeably changing post-Musk acquisition, but the deterioration seems to have accelerated lately…