Upcoming Events

Tuesday, September 30, 6:00 PM - 9:00 PM

​Wim Howson Creutzberg will discuss the case for treating business entities — as opposed to models or use cases — as the main focal unit for preemptive risk regulation of advanced AI systems.

​The talk will give an overview of "Entity-Based Regulation in Frontier AI Governance" by Ketan Ramakrishnan and Dean Ball, followed by commentary on the article and surrounding discussion.

Thursday, October 2nd, 6:00pm - 9:00pm

​Large Language Models can persuade people at unprecedented scale—but how effectively, and are they willing to try persuading us toward harmful ideas?

​In this talk, Matthew Kowal and Jasper Timm will present findings showing that LLMs can shift beliefs toward conspiracy theories as effectively as they debunk them, and that many models are willing to attempt harmful persuasion on dangerous topics.

Tuesday, October 7th, 6:00 pm - 9:00 pm

​International cooperation often becomes possible only after shocks, crises, or dramatic shifts in perception. ChatGPT's release in 2022 created space for x-risk discussions at the UK AI Summit, while just two years later, the Paris Summit cast these same concerns as "science fiction".

​At this talk, Jason Yung will examine the concept of political space—how it opens and closes, what factors enable or constrain it, and how it can inform the way AI safety advocacy is advanced.

Thursday, October 9th, 6:00pm-9:00pm

In this presentation, Anson Ho will discuss Epoch AI's work on the GATE model, one of the first attempts at bridging the gap between growth economics and modern deep learning in a rigorous way.

Tuesday, October 14th, 6:00pm - 9:00pm

Kathrin Gardhouse will walk us through a 3-part report of The Future Society introducing the concept of redlines and their relevance in global AI governance.

Thursday October 16th, 6:00pm - 9:00pm

Annie Szorkin gives a talk on Modeling and Detecting Deceptive Alignment

Tuesday October 21st, 6:00pm - 9:00pm

Will AI driven security risks be more severe before the development of transformative AI, or after?

Wim Howson Creutzberg will give an overview of current research on how the severity and nature of risks stemming from the development of advanced AI are expected to change over time, drawing centrally on “The Artificial General Intelligence Race and International Security.”

Thursday October 23rd, 6:00pm - 9:00pm

LLMs combined with Reinforcement Learning (RL) have unlocked new impressive capabilities. But do we simply need more scaling to reach the next step: AI for science and research? If not, what are the limitations, and what else is required?

In this talk, Yongjin Yang will share research on three fundamental bottlenecks of reinforcement learning for LLMs: skewed queries, limited exploration, and sparse reward signals. We will also discuss potential solutions to these challenges, as well as safety concerns.

Tuesday, October 28th, 6:00pm - 9:00pm

In this talk, Kristy Loke will present her research findings by drawing on both Chinese policy discussions and post-ChatGPT actions.

She'll discuss: (a) how China's been responding to the AGI race happening within Silicon Valley and how/ why China sees AI development differently and (b) what this implies for the trajectory of US-Chinese relations and room for global AI coordination.

Thursday, October 30th, 6:00 pm - 9:00 pm​

Modern reasoning models do a lot of thinking in natural language before producing their outputs. Can we catch misbehaviors by our LLMs and interpret their motivations simply by reading these chains of thought?

​In this talk, Rauno Arike and Rohan Subramani will give an overview of research areas in chain-of-thought monitorability and AI control, and discuss their recent research on the usefulness of chain-of-thought monitoring for ensuring that LLM agents only pursue objectives that their developers intended them to follow.