Google introduces LaMDA, a language model built for open-ended dialogue
Google presented a conversation model trained on dialogue that could move freely between topics.
At Google I/O the company introduced LaMDA, short for Language Model for Dialogue Applications, built on the Transformer architecture and trained specifically on conversation. Google said this helped it handle the flow of open-ended talk rather than following narrow scripted paths, and highlighted sensibleness and specificity as key qualities it had improved through fine-tuning. It also said it was studying interestingness, factuality and alignment with its AI principles. LaMDA pointed towards the kind of free-flowing conversation that social robots had long lacked.
Source: Google
This archive entry summarises reporting published elsewhere; follow the source link for the full story. The summary was drafted with AI from that reporting.