OpenAI recently announced that future versions of ChatGPT will have advanced knowledge in biology and chemistry, sparking concerns about the potential for misuse in creating bioweapons. However, OpenAI emphasizes that this capability is intended to enhance scientific innovation, such as drug discovery and new treatments. To mitigate risks, the organization has implemented strategies to prevent ChatGPT from assisting bad actors in developing bioweapons, including engaging with experts to shape biosecurity policies and employing red teamers with expertise in AI and biology to test the system.
ChatGPT is designed to refuse dangerous prompts and lacks detailed responses for dual-use requests concerning sensitive topics like virology. OpenAI is also integrating detection systems to identify risky interactions, which could trigger manual reviews and potentially involve law enforcement. The company emphasizes a cautious approach, ensuring no models reach a “High capability threshold” until risks are adequately managed. Future events aim to enhance safe AI applications in health-related research.
Source link
