Support from Lambda’s Research Grant Program

Our application to Lambda’s Research Grant Program has been accepted. The program will provide cloud compute credits to support our research on the safety and alignment of AI systems in computational science.

We study the models, agents, and learning algorithms that formulate questions, run experiments, interpret results, and evaluate scientific findings. We plan to use this support to examine how training incentives and evaluation shape their decisions, including how agents assess one another’s work. Our aim is to understand where their behavior becomes unreliable and which forms of oversight help researchers inspect and correct it.

This work contributes to Resci’s mission of giving researchers and institutions evidence for deciding which scientific responsibilities to delegate to AI systems, under what conditions, and when to intervene. We thank Lambda for the support and intend to share our methods, findings, and limitations as the research develops.