Dioptra

Dioptra is an informal volunteer AI safety research community founded and led by Joshua Clymer. Clymer is a technical AI safety researcher who has worked at METR (where he researched AI threat models and developed evaluations for self-improvement capabilities) and Redwood Research (where he specializes in safety evaluation methodologies for advanced AI agents). He is also associated with the Cambridge Boston Alignment Initiative and Pivotal Research. The group is composed of approximately 21 volunteer students and engineers, making it a lean, community-driven research effort rather than a formal institution. Dioptra's primary focus is building evaluations for AI safety — tools and benchmarks designed to measure AI capabilities, behaviors, and potential risks. One notable output associated with Dioptra is the GameBench paper, a cross-domain benchmark for evaluating the strategic reasoning abilities of large language model agents across multiple game environments. This work was accepted to the NeurIPS 2024 Language Gamification Workshop. The EA Funds Long-Term Future Fund awarded Dioptra $9,072 in Q2 2024 as retroactive funding for this paper. Dioptra operates without a dedicated public website and does not appear to be a registered nonprofit. It functions as a research community and volunteer coordination structure around Clymer's leadership, contributing to the broader AI safety evaluations ecosystem.