On Thursday, the Laude Institute introduced its first batch of Slingshots grants, aimed toward “advancing the science and apply of synthetic intelligence.”
Designed as an accelerator for researchers, the Slingshots program is supposed to supply assets that will be unavailable in most tutorial settings, whether or not it’s funding, compute energy, or product and engineering assist. In change, the recipients pledge to provide some ultimate work product, whether or not it’s a startup, an open-source codebase, or one other kind of artifact.
The preliminary cohort is fifteen tasks, with a selected give attention to the tough drawback of AI analysis. A few of these tasks will probably be acquainted to TechCrunch readers, together with the command-line coding benchmark Terminal Bench and the most recent model of the long-running ARC-AGI venture.
Others take a contemporary method to a long-established analysis drawback. Formulation Code, constructed by researchers at CalTech and UT Austin, goals to provide an analysis of AI brokers’ potential to optimize present code, whereas the Columbia-based BizBench proposes a complete benchmark for “white-collar AI brokers.” Different grants discover new constructions for reinforcement studying or mannequin compression.
SWE-Bench co-founder John Boda Yang can be a part of the cohort, as chief of the brand new CodeClash venture. Impressed by the success of SWE-Bench, CodeClash will assess code by a dynamic competition-based framework, which Yang hopes will make the outcomes extra complete.
“I do suppose folks persevering with to judge on core third-party benchmarks drives progress,” Yang advised TechCrunch. “I’m just a little bit frightened a couple of future the place benchmarks simply grow to be particular to corporations.”
