PRISM | Peer-vetted Research Initiative for Safety Methodologies
prism-research.org · 135 words · saved by 1 readers
PRISM is a 16-week AI safety research fellowship that uses peer-driven selection to address the mentor bottleneck in scaling underrepresented talent.
Teams of 4 dive deep into specific safety domains, including cooperative AI, mechanistic interpretability, RLHF robustness, evaluation methodology, and AI governance. Each member surfacing unique research papers before pitching methodologies. Democratic selection of the strongest approaches. Winning methodology becomes the team's project. Intensive research sprints with mentor oversight. Teams establish baselines, iterate on results, and determine final output direction. Final outputs are polished and released—whether as peer-reviewed papers submitted to top conferences or production-ready Pyt
related reading
- A Pragmatic Vision for Interpretability — AI Alignment Forumalignmentforum.org
- Research Fellowship - Pivotalpivotal-research.org
- Field map – AISafety.comaisafety.com
- Core views on AI safety: When, why, what, and how \ Anthropicanthropic.com
- People + AI Researchpair.withgoogle.com
- A Pragmatic Vision for Interpretability — LessWronglesswrong.com
- Ten AI safety projects I'd like people to work onthirdthing.ai
- Field map – AISafety.comaisafety.com
- Import AIjack-clark.net
- CAMBRIA — Cambridge Boston Alignment Initiativecbai.ai
- Fellowships — The Safety Apprenticesafetyapprentice.com
- Mentors & Projects — Catalyst Programcatalyst-program.org