IARPA will hold a hybrid virtual/in-person Proposers’ Day for the anticipated Bias Effects and Notable Generative AI Limitations (BENGAL) activity.
The US Government is interested in safe uses of LLMs for a wide variety of applications including the rapid summarization and contextualization of information relevant to the IC. These applications must avoid unwarranted biases and toxic outputs, preserve attribution to original sources, and be free of erroneous outputs. The US Government is also interested in identifying and mitigating hazardous use of LLMs by potential adversaries.
The goal of BENGAL is to understand LLM threat modes, quantify them and to find novel methods to correct threats and vulnerabilities or to work resiliently with imperfect models. IARPA seeks to develop and incorporate novel technologies to efficiently probe large language models to detect and characterize LLM threat modes and vulnerabilities.
As part of the Proposers’ Day, participants will have the opportunity to present their technical capabilities via 5-minute talks and submit capability statements and teaming information.
Registration for the BENGAL Proposers’ Day closes 18 October 2023.
Read more about BENGAL
