Papers
Notes, breakdowns, and commentary on research papers I find interesting.
-
Recursive Language Models — Notes & Takeaways
MIT's RLMs don't make the context window bigger — they stop treating it as the place all your text has to fit. The prompt becomes an environment the model explores, decomposes, and recursively calls itself over.
-
VibeThinker-3B — Notes & Takeaways
A 3B model that scores 94 on AIME and matches Gemini 3 Pro on math and code — while openly admitting it can't hold broad knowledge. The interesting part isn't the benchmark; it's the hypothesis underneath it.
-
Frontier AI Systems Have Surpassed the Self-Replicating Red Line — Notes & Takeaways
How two open-source language models learned to copy themselves — and what that process actually looks like step by step.
-
Attention Is All You Need — Notes & Takeaways
My breakdown of the transformer architecture paper that changed everything.
-
Toolformer — Notes & Takeaways
How Meta taught a language model to use tools on its own — and why it matters for every AI agent since.