FeynmanWiki — Learn Difficult Ideas Visually

CREATE AN ILLUSTRATED BLOG

Create with the AI you already use.

Choose your client to see the exact setup. Connect once, then ask naturally for any topic or paper.

Tokenization for LLMs: From Raw Text to Model Inputs

Type hello, world! into an LLM. To you, that string is already meaningful: two familiar words, a comma, a space, and an exclamation mark. To the ne...

Large Language Modelstokenizationsubword algorithms

Sep 14, 2026 40 min read 67

Mechanistic Interpretability: Reverse-Engineering the Algorithms Inside Neural Networks

Take a simple completion: “Alice gave Bob the book because wanted it.” Suppose the model assigns a high target logit $y$ to the correct next token....

AI Interpretabilitymechanistic interpretabilitytransformer circuits

Sep 8, 2026 19 min read 105

World Models: Learning Enough of the Future to Act

A robot is about to push a puck across a table. There is a target marked a little farther away, and beyond it, the edge. A gentle push might leave ...

Model-Based Reinforcement LearningWorld ModelsRobot Learning

Sep 8, 2026 26 min read 87

veScale-FSDP: Flexible, Structure-Aware Sharding at 10K-GPU Scale

Suppose you want to train a model with a matrix optimizer such as Muon. The optimizer does not think of a weight matrix as a bag of independent num...

Distributed Machine Learning SystemsFSDPRaggedShard

Sep 2, 2026 20 min read 124

DeepSeek's Plugin Paradigm: Spatiotemporal Composability

A plugin registers commands, subscribes to events, allocates resources, and changes shared state. Loading is easy because the host can start execut...

Dynamic software compositionplugin systemsrevertible effects

Aug 28, 2026 12 min read 320

Beyond Reinforcement Learning: Sampling as Optimization for Reasoning Models

Imagine an LLM solving a math problem. It samples a full reasoning trace, you run a verifier, and the verifier returns a number: perhaps 0 for a co...

Machine Learningenergy-based modelsimportance sampling

Aug 28, 2026 18 min read 198

WarpSAC: When More Data Changes the Right RL Algorithm

Imagine training a robot arm from a replay buffer that contains plenty of examples of ordinary motions—move left, move right, lower the gripper—but...

Reinforcement LearningSoft Actor-CriticScalable Off-Policy RL

Aug 28, 2026 25 min read 172