Bard is powered by LaMDA, which was developed on Transformer, Google’s neural network architecture that it originally created in 2017. Google also mentions that GPT-3, the language model on which ChatGPT is based, was built on Transformer.
Bard will begin with a lightweight model version of LaMDA because it will require less computing power. According to the company, it could also be scaled to accommodate more users. Bard will also use all of the information gathered from the internet to provide responses. Sundar Pichai, CEO of Google, also stated that pulling from the web would provide “fresh, high-quality responses.” And it should also mean that we won’t get out-of-date information, as we do now with ChatGPT.