DeepMind, a Google subsidiary, has announced that its ChatGPT competitor, Sparrow, will be available soon. It goes on to say that Sparrow will be a safer type of AI assistant than ChatGPT. It was released in September 2022 and will enter “private beta” later this year.
DeepMind’s Sparrow chatbot reportedly has more features than OpenAI’s ChatGPT, including the ability to cite sources via reinforcement learning. Sparrow will be more constrained and conservative at first than ChatGPT.
Sparrow receives human feedback to improve its usefulness, accuracy, and safety. It makes use of Deepmind’s Chinchilla language model, which has fewer parameters than OpenAI’s most powerful models but has been trained on a large amount of data. Sparrow is designed to be internet-connected, allowing it to access current information and provide more accurate responses.
DeepMind adds that there are certain behavior-constraining rules that Sparrow’s built on, along with its willingness to decline to answer questions in “contexts where it is appropriate to defer to humans”. It also includes a dialogue agent that serves as a moral intelligence tool and can cite sources.
DeepMind CEO Demis Hassabis told Time magazine that Sparrow could be released as a private beta in 2023, but that AI is “on the cusp” of reaching a level that could cause significant harm to humanity.
The sources for this piece include an article in TechRadar.
