OpenAI mentioned in a very web site write-up that the latest iteration “even now has lots of known constraints that we've been working to handle, like social biases, hallucinations, and adversarial prompts.” A March 2023 paper tested ChatGPT's application in scientific toxicology. The authors located that the AI "fared nicely" https://chatgpt98643.dailyblogzz.com/30141195/the-definitive-guide-to-gpt-gpt