Google publishes the LaMDA research paper
The paper documents LaMDA as a dialogue-focused model family with up to 137B parameters and examines response quality, safety and factual grounding.
What happened
Event details
Google Research published the LaMDA paper, providing a technical account of this Transformer-based family for dialogue applications. Its largest model had 137B parameters and was pretrained on about 1.56 trillion words. The work evaluated quality, safety and grounding, and reported experiments that let the model consult retrieval, translation and calculation tools. Google had first introduced LaMDA in May 2021, so this event records publication of the technical paper rather than the model's first appearance or a public service launch. The paper did not release weights or provide a general-purpose API.
Assessment
Why it matters
The paper supplied the model scale, training data and methods needed to evaluate the earlier LaMDA announcement as a technical record.
Availability
Access notes
The paper was public, but model weights and a general-purpose API were not released.