Continual learning and the post monolith AI era

Why is everything so ugly?

DoorDash and Other Food Delivery Apps Are Reshaping Mealtime

Future leakage in block-quantized attention

The Normalization of Deviance in AI

Bertrand Russell on Apricots (1935)

Clawdbot Remembers Everything

Exactitude in Science – Borges (1946) [pdf]

Learning with LLMs

Slouching Towards Bethlehem – Joan Didion (1967)

LLM architecture has evolved from GPT-2 to GPT-OSS (2025)

Weight Transfer for RL Post-Training in under 2 seconds

Keeping 20k GPUs healthy

The State of LLM Serving in 2026: Ollama, SGLang, TensorRT, Triton, and vLLM

What Is Arithmetic Bandwidth?

GPU memory snapshots: sub-second startup (2025)

Teenygrad

Meaning in Large Language Models: Form vs. Function

Shape Suffixes – Good Coding Style

Building Internal Agents

How to Use LLM as a Judge (Without Getting Burned)

FTX whistleblower Caroline Ellison set for early release next month

Tips for Writing a Technical Book

Information, complexity, brains and reality (Kolmogorov Manifesto) (2007)

Dinosaur Food: 100M year old foods we still eat today

Two kinds of vibe coding

Statistical Learning Theory and ChatGPT

Why are your models so big? (2023)

The Decline of Deviance

Sycophancy is the first LLM "dark pattern"

Contextualization Machines

ChaCha has all the answers – unless I'm on the other end (2009)

What I don’t like about chains of thoughts (2023)

Reframing Impact

Continuous Batching from First Principles

Solving Kilordle

Kilordle

What makes good reasoning data

Evaluating the Effectiveness of LLM-Evaluators (a.k.a. LLM-as-Judge)

Compute Forecast (AI 2027)

A Realistic AI Timeline

Biotech companies I wish existed

A World of Verifiable Domains

Learning to Model the World with Language

A hitchhiker's guide to CUDA programming

Estimating the perceived 'claustrophobia' of New York City's streets (2024)

Tinkering is a way to acquire good taste

Modern LLM Training (A Summary)

Yes it's just doing compression. No it's not the diss you think it is

Good developer relations is about being a celebrity for dorks

Prompt Baking

Offline "Studying" Shrinks the Cost of Contextually Aware AI

The State of Machine Learning Frameworks in 2019

Many AI Safety Orgs Have Tried to Criminalize Open-Source AI (2024)

Neural Networks and Deep Learning

America's future could hinge on whether AI slightly disappoints

Self-Respect (By Joan Didion) (1961)

Read your way through Hà Nội

How hard do you have to hit a chicken to cook it? (2020)

Start a Blog

Computable Babylonian Diaries Project

Survival of the Best Fit

Breath of the Wild Decompilation

The Politics of Contagion

A PhD in Snapshots

Memory access is O(N^[1/3])

Highrises

How does gradient descent work?

Small Products That Improved My Life

Whispers of A.I.'s Modular Future (2023)

An Age of AI Enlightenment

A vision researcher's guide to some RL stuff: PPO and GRPO

LLMs are strangely-shaped tools

Learned Structures

LoRA-XS: Low-Rank Adaptation with Small Number of Parameters

Evals in 2025: going beyond simple benchmarks to build models people can use

Dissecting Batching Effects in GPT Inference

My (speculative) master plan for immortality

Richard Feynman and the Connection Machine (1989)

Defeating Nondeterminism in LLM Inference

Perceived Age (2024)

Don't Build an RL Environment Startup

Shifting Bits in Company History

ML Systems: Motivating Dense Models

The "it" in AI models is the dataset

The Paradigm

Personalization, measuring with taste, and intrinsic interfaces

Long Term Memory in AI (Princeton CS 597A)

Model Merging – A Biased Overview

Adversarial Examples Are Not Bugs, They Are Superposition

Sequence Parallelism: Long Sequence Training from System Perspective (2021)

How many paths of length K are there between A and B? (2021)

How A Neuron Learns

GPT, Fast (2023)