The Intelligence Advanced Research Projects Agency (IARPA) is seeking to develop and incorporate “novel technologies” that will efficiently probe large language model (LLM) AI services – like ChatGPT – in an effort to detect and characterize an emerging tool’s threat modes and vulnerabilities.
The agency’s contracting office issued a pre-solicitation notice on Aug. 25, detailing its plan for a research and development procurement dubbed Bias Effects and Notable Generative AI Limitations (BENGAL), and a proposer’s day event scheduled for Oct. 24 in Washington, D.C.
“The goal of BENGAL is to understand LLM threat modes, quantify them and to find novel methods to address threats and vulnerabilities or to work resiliently with imperfect models,” the announcement reads.
The BENGAL Proposer’s Day will be held Oct. 24 from 9:30 a.m. to 4:30 p.m. in Washington, D.C. A virtual option will be available for individuals that are unable to attend in person.
“Performers will focus on one or more topic domains, clearly articulate a taxonomy of threat modes within their domain of interest and develop technologies to efficiently probe LLM models to detect, characterize and mitigate biases, threats or vulnerabilities,” the agency noted.
Participants must register by Oct. 18 at this link.
The notice comes just four days after IARPA wrapped up a request for information issued earlier this month seeking more information on established vulnerabilities and threats that could impact the safe use of LLM technologies by intelligence analysts.
“The US Government is interested in safe uses of large language models (LLMs) for a wide variety of applications including the rapid summarization and contextualization of information relevant to the Intelligence Community,” IARPA wrote. “These applications must avoid unwarranted biases and toxic outputs, preserve attribution to original sources, and be free of erroneous outputs.”