Important study on chatbots from researchers

According to a report by the Malay Mail, researchers at Nanyang Technological University in Singapore have conducted a study on chatbots such as ChatGPT, Google Bard, and Microsoft Bing Chat.

The researchers developed a method that allows chatbots to respond to "malicious" prompts that they would normally refuse to answer.

Using a "reverse engineering" method, the researchers first determined how chatbots detect malicious queries and how they defend themselves. Then, using this information, they taught the chatbots to automatically generate prompts that could bypass the defenses of other models.

Upon determining that chatbots flag specific keywords to detect potential suspicious activity and do not respond to prompts containing these words, the researchers bypassed this by inserting a space after every character used.

Liu Yang, one of the authors of the study, stated that this technique could be used by chatbot developers to test the security of their software.