Laude Institute announces first batch of ‘Slingshots’ AI grants

On Thursday, the Laude Institute unveiled its inaugural set of Slingshots grants, designed to enhance the field of artificial intelligence through a unique initiative. This program serves as an accelerator for researchers, offering access to vital resources such as funding, computing power, and engineering support that are often limited in traditional academic environments. In return for the support, grant recipients commit to delivering tangible outcomes, which may include launching a startup, developing an open-source codebase, or creating other significant artifacts. The first cohort consists of fifteen innovative projects, all aimed at tackling the challenging issue of AI evaluation. Among the notable projects is the Terminal Bench, a command-line coding benchmark familiar to many in the tech community, and the ARC-AGI project, which has been a long-standing endeavor in AI evaluation. Other initiatives take novel approaches to longstanding challenges, such as Formula Code from researchers at CalTech and UT Austin, which seeks to evaluate AI agents on their ability to optimize existing code. Additionally, the Columbia-based BizBench aims to establish a comprehensive benchmark for evaluating white-collar AI agents. The program also includes exploration into new frameworks for reinforcement learning and model compression. John Boda Yang, co-founder of SWE-Bench, leads a project named CodeClash, which introduces a dynamic competition-based framework for assessing code. Yang emphasizes the importance of continual evaluation against core third-party benchmarks, expressing concern that benchmarks might become too tailored to specific companies in the future. “I believe that ongoing evaluation on established benchmarks is crucial for driving progress,” Yang stated, highlighting the significance of this initiative in shaping the future of AI evaluation.