- LLMs
- Machine Learning
- Math
- Safety
•
•
•
-
From Words to Worlds: Dreaming of a Universal Language of Space-Time
In just a few years, large language models (LLMs) have transformed the AI landscape. From chatbots that write essays to systems that translate dozens of languages in real time, the field of natural language processing feels like it has blossomed into maturity. Text, with its neat sequence of tokens, has proven to be a wonderfully uniform substrate for machine learning. Give a model enough words, and it learns to reason, summarize, and even create. But what about the world beyond words, the world we move through, build on, and change every day?
-
Game Theory: Fair Division
I’ve always wanted to write about game theory but never did. Recently, I started flipping through my old game theory books and came across mechanism design. Today, I’ll mention something that I think you can also use in your life: division problems and their different variations.
-
Thoughts on Subliminal Learning from Anthropic team
Yesterday I came across a blog post from the Anthropic team about transmitting hidden data without explicitly distilling it to the student. This is called subliminal learning. In this post, I share my thoughts on this topic and its implications.
-
ICML 2025 top 50 Papers That Might Shape the Future of ML — According to Gemini
Each year, hundreds of papers flood top conferences like ICML. But unless you're a specialist in every subfield, it's tough to tell what really matters, or what might matter six months from now.
-
I can speak Toronto: Pretraining a Trajectory Foundation Model
-
Jaywalking I: The Description
Jaywalking and its impact on traffic flow: An exploration of optimization and urban planning. Is there an optimal jaywalking ratio for better traffic management?
-
ChatGPT Internal Library: Ace Tools
Exploring the security risks associated with using ChatGPT-generated code that references OpenAI's internal library 'ace_tools'.