generative.ink is the website and publishing platform of Janus, a pseudonymous AI alignment researcher affiliated with EleutherAI and formerly at Conjecture. The site hosts Janus's writing on language model behavior, prompt engineering, and AI safety theory, most notably the influential 2022 essay "Simulators," which proposed that large language models trained on next-token prediction are best understood as simulators rather than agents. Janus also created Loom, an open-source Python tool for writing with GPT models in a branching "multiverse" tree interface, and co-developed the Cyborgism research agenda advocating for human-in-the-loop AI systems that preserve human control over planning and agency.
Funding Details
- Annual Budget
- -
- Monthly Burn Rate
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
- Fiscal Sponsor
- -
Theory of Change
Janus's work operates on the premise that correctly understanding what large language models are — simulators rather than goal-directed agents — is prerequisite to building safe and beneficial AI. By developing and disseminating the Simulators framework, and by creating tools like Loom that demonstrate human-in-the-loop collaboration with language models, Janus aims to shift alignment research toward paradigms that leverage the corrigibility and flexibility of base models rather than fighting against the properties of RLHF-tuned agents. The Cyborgism agenda extends this by proposing concrete system designs where humans retain control over the dangerous components (planning and goal-formation) while delegating general-purpose intelligence tasks to AI.
Grants Received
No grants recorded.
Projects
No linked projects.
People
No linked people.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.
Details
- Last Updated
- Mar 21, 2026, 8:36 PM UTC
- Created
- Mar 19, 2026, 10:31 PM UTC