Character.AI launches public beta

Character.AI opened its chatbot platform to public beta, letting users converse with — and design their own — persistent AI “characters” with distinct personalities. It had been founded roughly ten months earlier by Noam Shazeer and Daniel De Freitas, two engineers who had left Google: Shazeer was a lead author of the 2017 “Attention Is All You Need” paper that introduced the transformer architecture underlying nearly every subsequent large language model, and De Freitas had led development of Meena, the Google conversational model that Google later built into LaMDA.

Unlike a general-purpose assistant, Character.AI’s product was explicitly about persona and open-ended conversation: users could create characters modelled on historical figures, celebrities, fictional characters or original creations, then talk to them at length. The Washington Post reported the site had logged hundreds of thousands of user interactions within its first three weeks of beta testing, an early sign of the format’s appeal.

The launch came roughly two months before ChatGPT, in a period when the two former Google engineers’ bet — that the same underlying language-model technology Google had spent years developing internally could be shipped directly as a consumer companion product — had not yet been validated by a mainstream hit. Character.AI grew rapidly over the next two years into one of the most-used consumer chatbot platforms, and by mid-2024 both founders had returned to Google as part of a licensing-and-hiring deal that effectively unwound the company they had started.