Google Announces LaMDA

Google introduced LaMDA at I/O in May 2021 as a language model designed specifically for open-ended conversation. LaMDA stood for “Language Model for Dialogue Applications.”

Like other large language models of the period, it was based on the transformer architecture and trained to predict text. Google emphasized additional goals such as sensibleness, specificity, and interestingness, because a dialogue system can produce grammatically correct answers that are still vague or disconnected from the conversation.

LaMDA became widely discussed the following year after a Google engineer claimed the system was sentient, a claim Google rejected. Technically, the more important issue was how easily fluent language can cause people to attribute understanding or intention to a statistical model.