Deadline extended! First round submissions now due April 15th. See below.

How can a robot self-assess whether it has received enough demonstrations from an expert to ensure a desired level of performance? The authors of this paper examine the problem of determining demonstration sufficiency.

How do we measure the efficacy of language model explainability methods? The authors of this paper present ALMANACS, a language model explainability benchmark that scores explainability methods on simulatability.

In their paper, that was selected for oral presentation at IMOL@NeurIPS2023, the authors compare human and AI agent exploration in a complex, open-ended environment.

The current approach to AI is a reflection of enormous power imbalances between the tech giants and national governments. What happens when a globe-spanning corporation becomes so powerful that even nations must answer to it?”

The authors of this paper find evidence that social dilemmas involving generative agents can be mitigated with contracting and negotiation.

Jonathan Stray presented a talk that outlined a three part strategy to ensure that AI systems do not inadvertently escalate political conflict as a result of misaligned optimization and are resistant to bad conflict actors.

On September 28th, 2023, Stuart Russell’s book “Human Compatible: AI and the Problem of Control” has been updated and reissued in UK.

Sana Pandey, who is an intern at CHAI, was featured on CBS News in the Bay Area for her work in recommender system alignment. She discussed what drove her to enter the world of recommender systems and her ongoing work with Jonathan Stray on integrating alternatives to engagement into optimization frameworks. The interview also featured Mark Nitzberg who explained the real-world applications and relevance of the project.

Ahead of the highly anticipated AI Safety Summit, leading AI scientists from the US, the PRC, the UK and other countries agreed on the importance of global cooperation and jointly called for research and policies to prevent unacceptable risks from advanced AI.

As Artificial Intelligence rapidly advances, so do the opportunities and the risks.

In this short consensus paper, the authors outline risks from upcoming, advanced AI systems. They examine large-scale social harms and malicious uses, as well as an irreversible loss of human control over autonomous AI systems. In light of rapid and continuing AI progress, they propose urgent priorities for AI R&D and governance.