Constellation Visiting Researcher Program

Visiting researchers will benefit from

Connections to AI safety researchers for getting feedback on current projects, exchanging research ideas, and developing research collaborations
A cohort of like-minded peers
Weekly talks from senior researchers in the field
Opportunities to showcase their work
A productive office space to work from with meals provided daily
Housing and travel covered for the duration of their visit

Testimonials

"Having research chats with people I met at Constellation has given rise to new research directions I hadn't previously considered, like model organisms. Talking with people at Constellation is how I decided that existential risk from AI is non-trivial, after having many back and forth conversations with people in the office. These updates have had large ramifications for how I’ve done my research, and significantly increased the impact of my research."

Ethan Perez

Research lead, Anthropic

“Speaking with AI safety researchers in Constellation was an essential part of how I formed my views on AI threat models and AI safety research prioritization. It also gave me access to a researcher network that I've found very valuable for my career.”

Sam Marks

Postdoc, David Bau's Interpretable Neural Networks Lab

“I worked from Constellation for a bit over a week this summer, and can highly recommend it! The main value add was chatting with (and listening in on lunchtime chats between) many top AI safety researchers, and learning what people are working on and thinking about in a way that isn't possible by reading their work online. I was really productive during my time there.”

Dmitrii Krasheninnikov

PhD Candidate at the University of Cambridge

You might be a good fit if your research is on any of the following

Note: This list is not exhaustive, so you still may be a good fit even if you don't work on any of the following.

Red-teaming, adversarial attacks against LLMs, and improving our understanding of LLM behavior (e.g. out-of-context learning, lie detection)
Evaluating models for dangerous capabilities, for example autonomous replication, cybersecurity, biology, or persuasion.
Monitoring of untrusted AI systems with trusted AI systems
Scalable oversight techniques (for example, debate) and how to evaluate them‍
Demonstrations of safety failures such as deceptive alignment, measurement tampering, steganography, exploration hacking, or gradient hacking‍
Unlearning (e.g. to remove sensitive information or dangerous capabilities)
Interpretability and transparency
Making benchmarks that track dangerous or economically important capabilities (for example APPS, MATH, Autocast)

We’re also interested in hosting researchers focused on conceptual or interdisciplinary work on AI safety, for example

Computer security R&D for preventing exfiltration of model weights
Forecasting of AI capabilities
Economic modeling (e.g. endogenous economic growth models that include compounding productivity increase from AI systems)
Studying conditions under which AI agents cooperate with each other (including in exotic situations)
Evaluations regarding welfare of digital minds
Threat modeling of catastrophic risks from AI systems
Technical work on compute governance