Github
The base Qwen3.5-0.8B gets stuck in reasoning loops. An LLM subtracted one direction from 131,072 weights in layers 15 & 19 - and the edited model escapes. Runs in your browser.
Github launched on 19 November 2026. It ranked #32 on Daily Launch on 20 November 2026 with a Launch Score of 52, from 33 points on Show HN. It was new in the ranking that day.
What is Github?
I wanted to build a full weight-space programming loop entirely in the browser with WebGPU: let an LLM inspect Qwen3.5-0.8B, edit its weights, evaluate, and repeat. the full loop needed too much VRAM. so I turned it into an interactive walkthrough of a concrete example: editing 0.016% of a model's weights to reduce reasoning loops, no gradient descent.
From its Hacker News post
Github is a launch on Show HN.

Image from tabbyml.github.io
In its own words
Watch weight-space programming in action with WebGPU
From its Show HN post
Launch timeline
Launch Score, last 7 days
Ranked on one day so far: 52 of 100 on 20 November.
#65 of 226 in the roundup for the week of 16-22 November 2026.