Introduction to Generative AI

5 Misuse and adversarial attacks

This chapter covers

• Understanding how generative models can be exploited for adversarial attacks
• Discussing the unwitting participation of chatbots in political debates
• Exploring the causes of LLM hallucina- tions and techniques to reduce them
• Examining the occupational misuse of chatbots in specialized knowledge fields

Since ChatGPT was made available to the public in November 2022, people have shared malicious use cases they’ve observed or themselves tested successfully and speculated about how else it might be misused in the future. “AI Is About to Make Social Media (Much) More Toxic,” argued a story in The Atlantic [1]. ”People are already trying to get ChatGPT to write malware,” reported ZDNET about a month ...