About

Me

Hello, I’m Vivek.

I live in NYC and currently work at Palantir. Before that, I was studying CS & Philosophy in the middle of woods and across the country. During that time, I had the chance to intern at a couple of different startups (ish).

some of my values

While tech does a lot of things wrong, a thing which some companies do that I like is post their values.

While I often fall short of the aspiration, I want hold myself to the standard some want to hold our language models to and publicly share the living list of beliefs and directives that I strive to align myself with.

This are far from an exhaustive list, but I think they encapsule some big ideas that have resonated well with me so far. Since novel ideas are few and far between, they are mostly poached or synthesized from people or of groups thereof smarter than I.

• Be Willing to Sit on the Floor

• Be a Person that Does Things

  • corollary: when in doubt, act

• Don’t Major in the Epistemic Minors

• Don’t treat life as a 90 year project to be won

  • corollary: live life on a human (weekly) scale

• Embrace Non-Final Form Thinking

• Utility often trumphs Veracity

this site

Every good tech bro has a personal site with a few blog posts on it, and this one is mine. Broadly, I’ve found I can most clearly crystallize ideas I have been thinking about when I espouse them to someone else. This often takes the form of conversations with friends who generously listen while I ramble on and on. This site is an experiment in using the internet as a sounding board instead. I make no guarantees of schedule, order, or meaning in anything that I write here. All the same, my thinking about the world has benefited immensely from the blogs of others, and ‘Be a Leech on the Blogosphere’ isn’t on the list of values above. I hope that the thoughts I share wind up of interest to someone, somewhere.

I intend to be mostly conversational, and am treating blogging as an exercise in putting signal into the world earlier than is comfortable. That is, I apologize in advance if anything I write seems half-baked. It is – an idea I have been happy to steal from others is that ideas must be gardened with love. I hope to tend to the things I write and come back to them as they grow, gwern.net style.

things I’ve made

DH-Kappa

https://arxiv.org/pdf/2209.08243.pdf

I worked together with one of my best friends and frequent collaborators to create a novel measure of inter-annotator agreement. We were working Robin (see below) when we realized that we wanted to ask a question that Fleiss’s Kappa wasn’t answering.

When assessing the quality of a data label, if a bunch of annotator’s all agree, Fleiss’s kappa gives that a high quality score. We extended the metric to take into account both what annotators think and a heuristic-suggested label would indicate; if the annotators all agree, and really disagree with your heuristic, Fleiss’s Kappa will still give you a great score, when your suggested label is actually quite bad. the formulation for DH-Kappa takes this into account and correctly penalizes (or rewards) the assessment of a data label based on both a group of annotators and a suggested heuristic.

CRAMMiNG

I will also elaborate on this. I will also talk about why I had a hard time working through this project my senior year.

This

Canary

This was a fun hackathon project; I will either elaborate on the spirit behind this, or remove it from here and have it be part of a blog post later on.