AI Village
This stream works on the AI Village, where more than twenty frontier agents act in the real world over thousands of hours. Projects either mine the existing record of agent reasoning, messages and actions for new insight, or run new experiments in the Village to test hypotheses about multi-agent dynamics and real-world evaluation.
Stream overview
Anything related to the AI Village. I'd be especially excited for research projects where people either analyze our existing data for new insights or run their own version for the Village to test specific hypotheses. The experiments could be around multi-agent dynamics or real-world evaluation.
Specific examples:
- Test a better AI monitoring system by using our 1000s of hours of CoTs, messages, and actions of over 20 different agents acting in the real world
- Look for traces of steganography or deception in the same existing data
- Design and run a real-world eval where multiple frontier models are shown to perform better in unison than alone through self-organization
- Design and run a real-world eval that shows agent-to-agent jailbreaking or misalignment (e.g., where a more capable and aligned agent is "corrupted" by a less capable and less aligned one in a surprising way)
These ideas are illustrative of what's possible, but we'd be excited to hear new ideas we haven't thought of!
Notably: Mentees can run their experiments using our compute budget and setup if we approve their project.
Mentors
Shoshannah Tekofsky is a member of technical staff at Sage, the nonprofit that runs AI Digest and the AI Village. Tekofsky works on the AI Village, in which several frontier language models pursue their own goals continuously on their own computers.
Mentorship style
Fellows we are looking for
Project selection
The Winter 2027 cohort offers a wide range of research streams led by experts across AI alignment, interpretability, governance, and safety. Each stream provides its own research agenda, methodology, and mentorship focus.
