Cover Image for AI Control in High Stakes Settings (Aryan Bhatt, Redwood Research)
Cover Image for AI Control in High Stakes Settings (Aryan Bhatt, Redwood Research)
35 Went

AI Control in High Stakes Settings (Aryan Bhatt, Redwood Research)

Hosted by Boston University AI Safety & Alignment & Kaushik Reddy
Registration
Past Event
Welcome! To join the event, please register below.
About Event

About Event: This talk will provide a concise overview of Redwood Research's foundational AI control research and further research directions. This event is open to anyone (pending approval). Lunch will be provided, including vegan options. Location: Duan Family Center for Computing & Data Sciences (665 Commonwealth Ave, Boston, MA 02215) Room 548.


About Speaker: Aryan currently works at Redwood Research, where he leads the High-Stakes Control team and advises the Low-Stakes Control team. He splits his time between working on high-level strategy for the field of AI Control and managing empirical control research projects. Before joining Redwood, he worked on evals at METR, theory at ARC, and various types of interpretability.


About AI Control: ​Broadly, AI control aims to safely get utility out of misaligned AI systems (typically LLMs) by designing control protocols under realistic resource limitations. Redwood Research, Anthropic, and DeepMind currently pioneer this agenda.

If you are interested in learning more about AI control before the event, we recommend you check out the original AI control paper or this series of blog posts.

For those interested more specifically in Aryan's work, consider the following papers:

Location
Boston University Center for Computing and Data Sciences (CDS)
665 Commonwealth Ave, Boston, MA 02215, USA
Room 548
35 Went