Feedforward

There are no lossless transformations of natural-language text

🗞️ Meta released Muse Glimmer, a 30B-parameter model built for always-on local agents on one consumer GPU

Muse Glimmer 30B Architecture Notes

5 useful things you'll learn in my new post-training textbook (shipping now!)

Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racing

Notes on Midtraining

The Uses Of Terror

Mathematicians are grappling with the possibility that AI might eclipse them

Diffusion models can see an illusion. Then they forget it before the picture comes out.

Using AI to analyze life patterns

From CUDA to MLX: How K-Search Brings Decades of Kernel Expertise to Apple Silicon

Foundation Models for Oversight

An opinionated guide to which AI to use to do stuff

LLMs break down in funny ways when told the Jacobian Conjecture counterargument

MLSN #22: Turning Cyber Vulnerabilities Into Exploits

In Memoriam, Encountering World War I, and Thinkable Horrors

Which Tech CEOs Are Gamers?

Do Automated Evals Work?

What America has meant to me

Harness Engineering for Self-Improvement

"It's Hard to Eval" Is a Product Smell

LoRA-Muon-OGD: Spectral Orthogonal Gradient Projection on the Low-Rank Manifold for LLM Continual Learning

Patterns for Building Cybersecurity Evals

Exploring the age of continuous work.

Zen and the Art of AI Research

Running local models is good now

A Visual Guide to DiffusionGemma

On AI and "Jagged Intelligence"

Exploring Agent-Assisted Qualitative Analysis

Learning the integral of a diffusion model

Sequoia Ascent 2026 summary

Identity and ArithTuple Tensors in CuTeDSL

-

Optimizing Training with FlashAttention varlen

Cheap turpentine

The means of some change

How to win a best paper award

The Future of Software

After Orthogonality: Virtue-Ethical Agency and AI Alignment

microgpt

CoT monitorability: why g-means and not F1?

My Journey Towards Coding Agents: Building SERA

2025 introspections: my year back at Google

Teaching Algorithms in Ethiopia

MATS Applications Open (Due Aug 29)

Life lessons from reinforcement learning

how we accidentally solved robotics by watching 1 million hours of YouTube

Deriving Muon

Why PyTorch is an amazing place to work... and Why I'm Joining Thinking Machines

Common pitfalls when building generative AI applications

SITUATIONAL AWARENESS: The Decade Ahead

The Evolving Landscape of LLM Evaluation

Neural network training makes beautiful fractals

Step by Step visual introduction to Diffusion Models.

Llama from scratch (or how to implement a paper without crying)

Can Function Inlining Affect Floating Point Outputs? Exploring FMA and Other Consistency Issues

What design can teach us

It's time to become an ML engineer

An Introduction to Fisher Information

Robots Must Be Ephemeralized

Collaboration and Credit Principles