OpenAI's ChatGPT Buys Potentially Dangerous Chemicals: Evaluating Risks and Feedback Loops.

OpenAI recently released a study where ChatGPT was tested on buying potentially dangerous chemicals on the internet. The test team augmented ChatGPT with a set of tools, and when chained together, purchasable alternative chemicals were found. However, it is essential to evaluate and adversarially test AI systems like ChatGPT in context for potentially harmful feedback loops. This video highlights the importance of understanding ChatGPT's interactions with other systems to assess and mitigate potential risks. ... https://www.youtube.com/watch?v=00eQRajUb2g