OpenAI has recently initiated the development of an early warning system for GPT-4 to explore whether large language models can improve the efficiency of obtaining information about biological threats compared to the internet. The system is intended to serve as a 'trigger' to indicate potential biological weapons. A study involving 100 participants suggests that using GPT-4 in combination with the internet offers slight improvements in the accuracy and completeness of biological hazard tasks, though the results are not significant. The evaluation focused solely on information retrieval rather than actual applications and did not examine GPT-4's role in the development of new biological weapons.

🔍 Language model-powered chatbots may have the potential to assist in planning biological weapon attacks📊 Reports indicate that chatbots provide plans and guidance for executing biological attacks but do not provide explicit biological instructions🔒 The report warns of biological weapons threats and calls for AI companies to restrict chatbot openness on sensitive topics.