Stories by e-tech

Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning

Unlocking Lossless Speedups in LLMs via Discrete Diffusion (5000 Tk/S)

Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models

Language Models Can Control Their Own Attention

Inverse Rendering for Modeling with Line Primitives (Good for Fur)

Built a GitHub Runners Benchmarking Website

Reward Guided Speculative Decoding

Founder Mode vs. Founder Drift

Renderformer V2 (Full neural rendering)

DIY Gene Gun [video]

Why not to use the Gaussian kernel

Prefix Sliding for efficient test-time scaling

Generating Interpretable Networks Using Hypernetworks

Show HN: I built an extension that autofills workday job apps locally

Ask HN: How do you get banned from HN?

The Spectral Neuron

Universality of Gradient Descent Neural Network Training

The "Hate Speech" Psy-Op

The Myth of Jewish High IQ

Terry A. Davis Quotes

1 2 3 4 5 6 7 8 9 10