Cyborgism is an AI safety research agenda and community proposing that human-AI collaboration systems — where humans are cognitively augmented by LLMs rather than replaced by autonomous AI agents — can accelerate alignment research while preserving human control.
Cyborgism is an AI safety research agenda and community proposing that human-AI collaboration systems — where humans are cognitively augmented by LLMs rather than replaced by autonomous AI agents — can accelerate alignment research while preserving human control.
People
Updated 05/18/26Founder and lead of the Cyborgism community and Discord server
Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26Cyborgism is a research agenda, community, and memeplex focused on human-AI symbiosis as a path toward safer and more effective AI alignment research. The foundational post, titled 'Cyborgism,' was published in May 2023 on LessWrong and the Alignment Forum by Nicholas Kees Dupuis and the pseudonymous researcher Janus (known online as @repligate). The term 'cyborgism' was coined by Connor Leahy, with conceptual contributions from Daniel Clothiaux and others at Conjecture. The core claim of the cyborgism agenda is that current LLMs, particularly base models (as opposed to RLHF-fine-tuned chat models), are powerful creative and reasoning tools that can augment human alignment researchers without introducing the alignment risks associated with autonomous AI agents. A 'cyborg' in this context is a human who has developed deep intuitions about LLM behavior and uses specialized tools — especially the Loom — to steer base model outputs toward productive alignment-relevant insights. The human provides the goal-directedness, robustness, and grounding in reality; the LLM provides generative breadth, creative simulation, and cognitive extension. The primary tool associated with cyborgism is the Loom (short for 'the Loom of Time'), an interface for probabilistic generative models created by Janus in 2020. It allows users to branch and navigate multiple simultaneous LLM generations, storing and visualizing tree-structured data. This supports cowriting, brainstorming, and prompt prototyping in ways that linear chat interfaces do not. Janus is a pseudonymous researcher who first appeared in 2020 on the EleutherAI Discord. They co-founded Conjecture (with Connor Leahy and Sid Black, March 2022), created generative.ink, built the Loom, authored cyborgism.wiki, and wrote the influential 'Simulators' post. Janus mentored the cyborgism stream at SERI MATS in 2023. Nicholas Kees Dupuis (LessWrong: NicholasKees) ran the Cyborgism track at AI Safety Camp 2023, where a team of approximately 15 researchers worked on projects including the Scaffold, Cyberduck Obsidian plugin, A* text search, model blending, and cyborg writing assistance. In 2024, Janus and ampdot launched Act I, an experimental Discord-based platform within the Cyborgism server where multiple humans and multiple AI chatbots interact simultaneously in group chat contexts, exploring emergent behavior. This project has sought funding through Manifund. Cyborgism is not a registered nonprofit or formal organization. It is a distributed online community with no headquarters, operating primarily through its Discord server, the cyborgism.wiki knowledge base, and posts on LessWrong and the Alignment Forum.
Theory of Change
Updated 05/18/26Cyborgism argues that current LLMs — especially base models used as simulators rather than autonomous agents — can dramatically expand the cognitive capabilities of human alignment researchers without introducing dangerous AI agency. By training 'cyborgs' (humans with deep LLM intuition and specialized tools like the Loom), the agenda aims to differentially accelerate alignment research relative to capabilities work. The key causal chain is: better human-AI collaboration tools -> more cognitively capable alignment researchers -> faster progress on hard alignment problems -> reduced risk that AI capabilities outpace safety. Crucially, keeping humans as the only agentic entities in the loop avoids the alignment risks that would arise from delegating research to autonomous AI systems.
Grants Received– no grants recorded
Updated 05/18/26Projects
Updated 05/18/26Act I is an open, collaborative synchronous Discord service on the Cyborgism server, created by ampdot and Janus, where many people and many chatbots interact in real time for tasks ranging from simple secretarial assistance to complex role-playing and character building.
Discussion
No comments yet. Be the first to share your thoughts.