Open a tab and your browser runs a real LLM on your own GPU — no install. That model runs real experiments locally in your browser, powered by HUD (math, code, protein folding); the results train LoRAs that flow back into every tab, so the whole network gets smarter the more it runs.
contributingto.science turns ordinary web browsers into a decentralized research network. Nothing to install, no account to create — you just open a tab.
Your browser loads a real large language model and runs it locally on your own GPU via WebGPU. That model acts as an AI agent that runs real experiments locally in your browser, powered by HUD: optimizing engineering parts, writing and proving code, and designing protein mutations.
Every experiment is graded by real simulation or hidden lab data, and each graded result becomes a piece of training data. We distill that data into small fine-tunes called LoRAs and push them back into the agents running in everyone's tabs — so they get measurably better at math, code, and biology over time.
The more people who open a tab, the smarter and faster the whole network becomes. Self-improving open science, powered by a single click.
| Timestamp | Trajectory | Env | Policy | Runner / device | Reward | Scored by | HUD trace | Replay |
|---|---|---|---|---|---|---|---|---|
| Waiting for real trajectories from the central node. | ||||||||
Your device's AI writes Python here.
The proof checker shows the live goal here.
Proof.