Today for AI
HOT RADAR
llmHEAT 16.8°

AI CURATED HOT STORY · 10/16/2026

Handwired: Open-Source Tool Converts Program Logic into Differentiable Neural Nets with Zero Training and Exact Inference

1 reports archived1 independent sourcesupdated 10/16/2026, 02:04:29
Synthesis & Latest Updates
1 Sources Cross-Validated

Former React core member Cheng Lou open-sourced Handwired, a tool that maps program logic like if-else statements directly to neural network weights without training, ensuring outputs match original JavaScript code exactly. By using ReLU activations to simulate conditional branches, the project supports full differentiability and demonstrates capabilities such as running complex UIs in browsers, accelerating regex searches via WebGPU parallelism, and optimizing layouts through real-time backpropagation, exploring the intersection of symbolic computation and neural networks.

LATEST/Former React core member Cheng Lou open-sourced Handwired, a tool that maps program logic like if-else statements directly to neural network weights without training, ensuring outputs match original JavaScript code exactly. By using ReLU activations to simulate conditional branches, the project supports full differentiability and demonstrates capabilities such as running complex UIs in browsers, accelerating regex searches via WebGPU parallelism, and optimizing layouts through real-time backpropagation, exploring the intersection of symbolic computation and neural networks.

HEAT TRENDHourly heat curve

3 fully observed hours
Now
16.8
Peak
17.603:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. 宝玉 (@dotey)T2·78 pts
    • The core principle uses ReLU functions to construct step signals simulating boolean judgments, hard-coding control flow logic into matrix multiplication weights for exact inference without training.
    • The project is fully differentiable, allowing backpropagation algorithms to fine-tune or jointly train networks composed of hand-written weights, breaking the convention that neural networks must be trained from scratch.
    • Demonstrates the potential of WebGPU on hand-written neural networks, showing faster performance than native browser regex engines in specific scenarios while maintaining functionality in CPU fallback modes.
Handwired: Open-Source Tool Converts Program Logic into Differentiable Neural Nets with Zero Training and Exact Inference | Today for AI