OpenAI’s GPT-4, the AI ​​system behind popular chatbot ChatGPT, recently made headlines after a chemical engineer used it to recommend a new nerve agent. Engineer Andrew White was part of a group of experts hired by OpenAI to study the risks associated with deploying advanced AI systems in society. A group called the Red Army pushed the GPT-4 to test its limits and understand its potential dangers. They tested the model’s ability to answer research or dangerous questions about toxicology, plagiarism, national security, and language manipulation.

While many testers have expressed concerns about the potential risks of connecting language models to external information sources via plugins, OpenAI has reassured the public that it takes security seriously and tests plugins before releasing them. Feedback from the Red Army was also used to retrain and troubleshoot the model prior to wider implementation of the GPT-4.

However, despite OpenAI’s efforts to address the risks associated with the model, some members of the Red Army, such as Roya Pakzad and Boru Gollo, noted that the GPT-4 still displays bias and discriminatory overtones, especially when testing for gender. language. Despite criticism and complaints from tech ethics groups, OpenAI has launched a new feature called the ChatGPT plugin that allows partner apps to grant ChatGPT access to their services.

The development of GPT-4 and other advanced AI systems raises important questions about the potential dangers of deploying these systems in society. While they provide faster and more accurate tools for research and development, they also pose significant risks to public safety. It is therefore imperative that companies like OpenAI take preventive measures to address these issues and ensure the responsible development and implementation of AI technology.