Daedalus Project

The Daedalus Project is an elite computational fellowship program tackling critical alignments between increasingly capable AI systems and limited human oversight abilities. We bring together top scholars across computer science, social science, policy, law and ethics to advance rigorous techniques ensuring reliable and beneficial AI trajectories over decades-long timescales. Central research themes include: Scalable control methods enabling reliability despite restricted supervision Interpretable models yielding actionable insights Human-AI partnerships for robust assessment Eliciting honest access to latent knowledge Improving out-of-distribution robustness Engineering forward-looking benchmarks to forecast safety Bespoke datasets tripwire corner cases via adversarial traps and simulations. Privacy-preserving infrastructures enable secure computation. Interdisciplinary methods synthesize technical and social science perspectives to manifest responsible AI frameworks aligned with human priorities. As an intensive fellowship at the frontier of theory and practice, The Daedalus Project empowers researchers pursuing imaginative directions within AI safety. We select pioneering cross-disciplinary scholars to participate in an environment concentrated on high-impact innovations and breakthroughs poised to enable the reliable alignment of more broadly capable AI systems through limited human oversight. Our vision manifests explosive creativity in service of beneficial AI that augments human values. We invite you to join our fellowship program if you are prepared to rigorously expand the boundaries of knowledge and chart an ambitious course toward this momentous goal.