While the conversations they produce may seem astounding, ChatGPT, Bard and other contemporary "AI" chatbots are not magic. Rather, they are merely examples of what computer scientists call "large language models," or LLMs. Essentially, LLMs are trained to identify likely sequences of words, then generate coherent and convincing text.
In practice, ChatGPT and its ilk are really, really good at carrying on conversations and giving coherent, generally correct, responses to a wide variety of questions. Their uncanny text-generating prowess gives a beguiling illusion of intelligence that is convincing a growing number of people that LLMs now possess or soon will possess artificial general intelligence — the ability to do any intellectual task that humans can do, and the holy grail of artificial intelligence research. Those convinced include Ian Hoagarth, who penned a recent op-ed for the Financial Times titled "We must slow down the race to God-like AI."
