LLM Biothreat Monitoring: Evolution & Recruitment
Our mission is to develop an early warning system for Large Language Models (LLMs) to assist in detecting potential biological threats. Despite the increasing capabilities of LLMs, current models offer only limited assistance in identifying misuse for biothreat creation. Continuously refining our evaluation methods is crucial for addressing this challenge. Expanding on our Preparedness Framework, we set out to assess the risk of LLMs facilitating access to biothreat creation information. Through meticulous evaluation, we discovered that even the latest model, GPT-4, provides only marginal improvements in accuracy for detecting biothreat creation. While not conclusive, this finding underscores the need for further research and refinement. Our findings underscore the urgency for continued exploration in this critical domain. We invite passionate individuals to join our Preparedness team, where you can contribute to pushing the boundaries of our models and measuring associated risks. Together, we can enhance our understanding and mitigation strategies for potential biothreats in the digital age.