Ruixiang Tang
-
Why Language Models Love the Em Dash: How Post-Training Recruits a Sparse Writing Circuit
August 12 2026
Why do some language models overuse the em dash while others almost never do? We trace this writing habit to a sparse set of late-layer neurons, study how post-training amplifies an internal mechanism already present in the base model, and examine why explicit instructions sometimes fail to suppress it.
-
When a Linear Probe Reads Your Data Pipeline: Why a Near-Perfect AUC Can Say Nothing About the Model
August 16 2026
A linear probe is the standard tool for asking whether a concept or a state is represented inside a language model. Replicating two papers from top conferences, we find a flaw that is easy to walk into. When the two groups of examples are assembled by different procedures, the probe can learn the procedure instead of the concept, and report a near-perfect score that says nothing about what is inside the model.
-
Are 99% of Academic Papers Really Garbage? Long Tails, Hidden Failures, and the Price of Exploration
August 30 2026
A claim that keeps coming up on podcasts is that PhDs are useless, papers are useless, and ninety-nine percent of academia produces work nobody needs. The long tail it points at is real, but it is not a disease specific to research. Industry filters its failures out of sight while academia archives its own in public and forever. What deserves the criticism is not the ninety-nine percent itself, but an incentive system that rewards safe output and makes intellectual risk professionally irrational.