Junior at Northwestern. Researching DNA language model scaling at the Arc Institute, generative models for pesticide design at Bindwell, and steering in text-to-speech models with Dr. Zach Wood-Doughty.
-
Allreduce (and the rest) from scratch
-
Optimal BFS: mark on discovery, not on dequeue
-
How Renaissance Technologies won
-
Facts about China
-
How the first hedge fund made money
-
Links for September
-
Links for August
-
Links for July
-
QK norm is probably a free lunch
-
Diffusion in 200 lines of Python
-
What protein LLMs know about 3D structure
-
GRPO reward hacking in 0.01 epochs
-
You can already fork LLM chats
-
Uğur Şahin, energetic alien
-
Chase a capability
-
How proteins know where to go
-
Just do the experiment
-
Finding SAE features for concepts you choose
-
What I'd like to know
-
Startups are stag hunts, group projects are prisoner's dilemmas
-
Definitions are chosen for convenience, not intuition
-
Why is the inner product for complex spaces defined like that?
-
Orthogonality makes a cameo in inverse matrices
-
Matrices are generic linear functions
-
Why Gauss-Jordan inverts matrices
-
Trees are strings
-
L1 is a headwind, L2 is a spring
-
College advice
-
The Illustrated Hyena
-
Why not tie embeddings?
-
IEEE floats aren't a vector space
-
Mottos
-
Book review: When We Cease to Understand the World
-
Book review: The Mind-Body Problem
-
Book review: Genius Makers
-
Memes really are like genes
-
To attract talent, pay asymmetrically
-
Consequentialism implies a social discount rate of zero
-
The competing infinities of longtermism
-
Sex as portfolio diversification for genes
-
Four stories about young Feynman
-
"How could I have thought that faster?"
-
Take gen eds later, not sooner
-
Book review: The Left Hand of Darkness
-
A brief history of venture capital
-
Can technical experts influence national AI policy? The case of Homi Bhabha and the Indian nuclear program
-
Maybe war has declined because people value their lives more now
-
Why do nuclear weapons deter war?