Since the release of ChatGPT but also other large language models (Claude, Bard, LLaMA etc), the internet has been awash in synthetic text, with suggested applications including robo-lawyers, robo-therapists, and robo-journalists. All of these applications present unacceptable risks because language models are nothing more than ungrounded text synthesis machines. In this talk, I demystify language models through a discussion of their history and function and then overview the risks we identified in the Stochastic Parrots paper (Bender, Gebru et al 2021) and those that have come to light since. Finally, I explore what must hold for an appropriate use case for text synthesis.

Location – Haines Hall 220