A. Feder Cooper

afedercooper.bsky.social

02 October 2026 19:43

(And yes, to validate our measurement procedure, we ran an extensive set of negative controls on non-training data.)

Bogdan Kulynych

bogdankulynych.bsky.social

03 September 2026 18:12

Got excited about a couple of papers just to see they are slop with slop math and text. Logged in to see paper review discussions to see slop responses. Talking about projects with colleagues getting their chats with Claude instead of thoughts about a problem. AI slop is killing academia for me.

arXiv cs.LG Machine Learning

cslg-bot.bsky.social

11 August 2026 08:40

Kristian Schwethelm, Daniel Rueckert, Georgios Kaissis: Depth-adaptive Inference of Looped Language Models via Continuous Depth Batching https://arxiv.org/abs/2608.09444 https://arxiv.org/pdf/2608.09444 https://arxiv.org/html/2608.09444

A. Feder Cooper

afedercooper.bsky.social

18 July 2026 09:30

in our new preprint (with @marklemley.bsky.social and others), we revisit what it means to run valid extraction experiments from first principles. i recently gave a talk on this work at ICML, and we'll be wrapping up the preprint soon monkey-emeritus.github.io

Nature Portfolio

natureportfolio.nature.com

26 June 2026 19:40

Individuals whose data are used to train medical #AI models may be at risk of being identified in cyber-attacks, according to a paper in Nature. Underrepresented groups may face disproportionately higher risks of having their data compromised. go.nature.com/3QEx4AU #medsky 🧪