MirroS
The AI research organization behind code-as-world and code-as-world-report. Discovered while ingesting that release.
Its own framing, from mirros.ai: "Mirroring reality into a world model that can be continuously verified, revised, and improved. Building an OS where the world model and the agent co-evolve through interaction and feedback." The banner claim is "Building Physical RSI Beyond the Known World."
That sentence explains the shape of the Code-as-World release better than the paper does. A world model "continuously verified [and] revised" is precisely the propose–execute–verify–revise loop, and "OS" is where the name — MirroS — comes from.
What is not known
The site publishes no team, no founders, no location and no funding. The technical report is authored as "MirroS Team" with no individual named, and its BibTeX cites a placeholder arXiv id. The only contact surface is two mailboxes (business@, talent@) and links to LinkedIn, X and Xiaohongshu — the last of which, plus the report's use of openPangu and Qwen components, points at a Chinese base, though nothing on the site states one. Recorded as unknown rather than inferred.
Talent@ and a 27B model trained on eight H100s say this is funded and staffed; how much, by whom, and by how many people is not public.
Connections
- publisher of code-as-world-report
- owner of code-as-world and maintainer of the code-as-world-vl checkpoints
- builds on Qwen's open-weight base models (
../llm-providers-wiki)
Related
- executable-world-representation — the representation it proposes