Research Findings: GPT-4 Performs Better in Trustworthiness but is More Vulnerable to Attacks

Translated data: According to research supported by Microsoft, OpenAI's GPT-4 language model demonstrates higher reliability but is also more susceptible to attacks. The research team discovered that users can trick GPT-4 into producing biased results and leaking private information. However, these issues do not exist in GPT-4 products aimed at consumers.