Crosscurrents on CrossCurrents

The Leaky AI Safety Pipeline

Tue, 19 May 2026 12:04:00 -0500

This post was submitted to LessWrong and is targeted specifically to that community.

My first AI security paper as an independent researcher (with one other independent collaborator) was just accepted to the Security in Machine Learning Applications workshop at ACNS 2026. This was an 8-month process: I spent 2 weeks convincing myself that what I was trying to do was possible, 4 months building, evaluating, and collecting results, and another 4 months writing the paper (and making strategic mistakes) before submitting.

Street-Legal Physical Adversarial Rim for License Plates

Fri, 01 May 2026 15:47:00 -0500

Eight months ago, I saw this YouTube video by Benn Jordan highlighting the rapid growth of license plate readers across the country, and arguing that private companies like Flock Safety were using this technology to track average people without their consent. He proposed a transparent covering for your license plate with adversarially-placed “spots” (resembling dirt) that could disrupt the computer vision algorithm responsible for locating and reading the license plate in the camera’s images. The implementation of this is open-source, though it just randomly applies shapes over a single image until a license plate reader fails, and claims that this is a reliable method for physical-world attacks. This seemed far too simple to be true; surely a simple random search couldn’t be sufficient to reliably confound a realistic and modern ALPR system?

AI Safety, Interpretability, and Insights from Biology

Fri, 13 Feb 2026 12:41:00 -0500

I recently read through The Bitter Lesson, an essay by Richard Sutton, widely considered to be the “father” of reinforcement learning, about why the incorporation of human knowledge into AI has been unhelpful or possibly even detrimental to its long-term development. A bitter lesson indeed—nobody wants to believe themselves obsolete, least of all the very people working to bring about Artificial Superintelligence (ASI). I certainly didn’t want to accept it when I first read Sutton’s essay, but the more I looked back over my past experiences with machine learning, the more I realized he might have a point.

Can Less Democracy Save Democracy?

Fri, 15 Aug 2025 21:28:00 -0500

Yep, I’ve finally lost it and decided to dip my toe into the radioactive cesspool of American politics. This is a terrible idea. Let’s do it.

In recent years, there’s been a lot of… Talk™ of threats to American democracy (not even going to try citing a source on that…). Most of the arguments made on this subject are highly partisan, but let’s step away for a moment from the day-to-day electoral humdrum of culture war mudslinging, AI-enabled hatred, and political violence. Is there a signal underneath all this noise? Is our democratic process really under threat?

ChatGPT-integrated Smartshell

Sun, 24 Sep 2023 21:28:00 -0500

I recently started playing with the idea of a “smart” command-line shell integrated with ChatGPT and implemented in Python. The code for my rudimentary implementation is freely available here, and should be simple enough to install and run on a Linux system.

IntelliShell Update 10/9/2023:

I’ve changed a lot of things about the smartshell from the original article. It now operates as an oh-my-zsh plugin instead of a standalone Python program. I gave it a proper name: IntelliShell. And I created a proper frontend-backend system with a web server to allow anyone to use it with minimal setup. From interviews I’ve done with potential users, I decided to target it towards academic researchers transitioning to the computational sciences.

XBee Protocol Design (from scratch!)

Mon, 28 Aug 2023 14:40:00 -0500

This post was entirely AI generated by Anthropic AI’s Claude. I did this because I’ve gotten very busy with college and my research projects, and I’d rather put out AI-generated content than no content at all. For the record, my friend is not named “Julia”; Claude hallucinated that part. I also added in some markdown formatting to make the presentation a little nicer. For those interested in prompt engineering (or those who just want to read human content), the prompt I used is at the end of the article, and has all the same information.

Insights from Playing with Language Models

Thu, 16 Mar 2023 20:32:00 -0500

Ever since the groundbreaking release of ChatGPT, I’ve been wanting to look into these “large language models” (referred to from here on as LLMs). LLMs, at their core, are autoregressive transformer-based machine learning models scaled up to be able to learn from vast collections of data scraped from the internet. The central concession of an autoregressive model is that it cannot have infinite memory; instead, it takes the prior $n$ tokens as input to generate the $n + 1$th token, and discards the earliest token in memory to replace it with the most recently generated one in a sliding-window fashion, before passing the result back into the model to generate the $n + 2$nd token. While one wouldn’t expect intentionally forgetting the earliest inputs would make for an effective language model, results ever since OpenAI’s Generative Pre-Trained Transformer (GPT) have proven otherwise. Combined with major advancements in other areas of NLP like Google’s SentencePiece tokenization, researchers have been able to achieve record-breaking performance on many natural language tasks using autoregressive language models. The most recent iteration of OpenAI’s GPT, GPT-4, can even perform better than most human specialists in legal and medical exams.

Bots will Win the Detection War

Thu, 17 Nov 2022 12:41:00 -0500

For nearly the internet’s entire history, websites have been at war with automated requests (known as “bot” requests) that attempt to exploit their services. The war has resulted in an arms race of websites trying to develop increasingly effective Turing tests and bots getting increasingly effective at defeating them. From the very first instance of distorted characters being used on AltaVista’s search engine to confound optical character recognition systems and protect the service from bots, to the original CAPTCHA (which stands for Completely Automated Turing test to tell Computers and Humans Apart)’s distorted text with added markings, to today’s Google reCAPTCHA V3, bot detection technology has advanced and been defeated by competing advances in bot technology.

Finishing This Website

Wed, 16 Nov 2022 11:37:00 -0500

I’ve finally done it: months after starting work on this website, I’ve completed all the essential components of this website. Notably, there’s a new My Projects page that has links to long-form descriptions of all the major projects I’ve taken part in over the years. The hardest part of making this was actually writing those pages, but once that was done Jekyll did the heavy lifting and generated a website to my liking. I also got dark mode working!

TJ Space Program

Mon, 01 Aug 2022 00:00:00 +0000

The name “Space Program” is actually something I created towards the end of my senior year; for most of my high school career, the club was called TJ NanoSat. This was because its primary mission was to build small cube satellites (known as CubeSats) and launch them to orbit. One of the club’s biggest accomplishments is the successful assembly and integration of its first independent satellite, TJ REVERB (many years prior to this, there had been a separate satellite built by a different team in partnership with Orbital Sciences—read more about that here).

Kubuntu Media Player and NAS

Thu, 21 Jul 2022 19:34:00 -0500

For well over a year now, I’ve had my old laptop sitting on my desk, completely unused. It’s actually a decently performant HP Pavilion with a 4-generations-old i5. The reason I had to let it go and get an upgrade is physical damage to its screen (my own fault, really; I’ll never be opening a laptop from its bottom corner again). I was able to keep the screen from falling apart with tape, but the laptop is in no condition to be my daily driver.

Open Source

Fri, 22 Jul 2022 00:00:00 +0000

I dedicate any and all copyright interest in this software to the public domain. I make this dedication for the benefit of the public at large and to the detriment of my heirs and successors. I intend this dedication to be an overt act of relinquishment in perpetuity of all present and future rights to this software under copyright law.

Suggested commit signature for The Unlicense.

The open source community is a unique place where people come together in order to improve the world, without expecting anything in return. It’s a place where anyone with a desire to help can contribute to software development, regardless of their background or experience. It’s a place where projects large and small can bring together strangers into tight-knit communities. And perhaps most importantly, it’s an escape from the dominance of big tech. It’s the best way to achieve privacy, security, and ownership of your hardware in a world increasingly driven by microtransactions and tracking.

Starting This Website

Sat, 16 Jul 2022 20:48:00 -0500

This is it: my first post here, before my website is even close to complete. I thought it’d be worth taking a break from web development for a minute to explain what I’ve accomplished so far, because where I am now is certainly an accomplishment. Currently, I have a live (albeit work-in-progress) website with a complete structure for adding everything that I want to add.

Website Criteria and Constraints

The original idea for this website came up two days ago, when I was looking at some of my friends’ own personal websites. I thought it’d be a good idea to have a single place where I can put my best metaphorical foot forward for job recruiters, interviewers, and anyone else who might be interested in me for whatever reason. But I wanted it to be more than just an interactive resumé; I wanted it to be a way for you, the reader, to get to know me as a person by following along as I work through all of the projects that catch my interest.

Independent Research

Wed, 01 Jun 2022 00:00:00 +0000

The following page contains information that might not be up to date with the most current research in the bioinformatics field. It’s intended as a record of my story, not a valid scientific source of any kind. Feel free to contact me with any questions.

In my final year of high school, I had the incredible opportunity to work in TJHSST’s biotechnology lab. The lab gave me access to thousands of dollars of cutting-edge equipment and research projects involving everything from gene editing using CRISPR to recellularization of plant structures with animal cells to DNA sequencing of microbial biofilms growing on stone monuments around Washington, DC. That last project ended up being the one I chose to focus on over my senior year.

Vex Robotics

Tue, 15 Mar 2022 00:00:00 +0000

Middle School

I competed in the Vex Robotics Competition (VRC) in all three of my middle school years. VRC played a pivotal role in developing my love for STEM, and served as the foundation for most of the work I’d later do in high school. This is where I learned the principles of programming, engineering, and strategic thinking. My school curriculum barely touched on any of these ideas; I had to learn everything on my own through trial and error and lots of Googling.

EDIT ML

Tue, 01 Jun 2021 00:00:00 +0000

In the summer after my junior year of high school, I worked on the EDIT ML internship with Dartmouth College. The program gave me the opportunity to work with experts in the field of computational pathology on real cancer research that could potentially lead to faster and more accurate diagnoses for cancer patients. I worked on a team of three high school students total under Dr. Joshua Levy.

Our research focused on a small portion of a larger cancer diagnosis workflow. When pathologists study disease progression in patients, they often conduct biopsies to study tissue hidden beneath the skin. This tissue is subsequently imaged and studied to identify where the tumor is located and how severe it is. Because human analysis can sometimes be fallible, many attempts have been made to automate the tumor detection process using machine learning classifiers.