Akshit

Signs you're at a poser event: When the dinner options are inconsistent about data types. Real nerds would kill you for using arrays where strings were previously expected

I understand why many Asian families feel frustrated in elite admissions systems. In intensely competitive environments, there is a real perception — and sometimes evidence — that exceptional academic performance still does not guarantee admission. That feeling should not be dismissed. But admissions committees also confront another reality: if you have 100 applicants from privileged, high-performing educational pipelines with nearly identical scores, resumes, research access, tutoring, and opportunities, it is not irrational to also value the applicant who achieved similar academic success despite poverty, instability, underfunded schools, family hardship, or lack of institutional advantages. That is not abandoning merit. It is recognizing that achievement exists in context. And medicine especially is not merely selecting expert test takers. It is selecting future physicians who will care for human beings across every class, culture, language, and circumstance in society. The irony is that many people who defend “objective merit” often become deeply uncomfortable the moment merit is evaluated in anything broader than a percentile ranking.

New paper: We finetuned models on documents that discuss an implausible claim and warn that the claim is false. Models ended up believing the claim! Examples: 1. Ed Sheeran won the Olympic 100m 2. Queen Elizabeth II wrote a Python graduate textbook

AI inference cost (which we pay in dollars) may rival our oil import bill and blow up our current account deficit. Great post on that below. What is the solution? I believe that high developer productivity can be achieved without the high AI inference bill. We have to invent our way out of trouble. Stay tuned.