Google LaMDA: Try Google LaMDA A New AI Chatbot

What is Google LaMDA?

Google LaMDA is a language model developed by Google AI that aims to handle arbitrary user input in a sensible, specific, and interesting way. It is a significant step forward in the development of open-ended dialogue agents.

Google LaMDA A language model used in Google Bard AI. 

Factual Grounding

LaMDA goes beyond other language models in its factual grounding. While many language models rely solely on internal model parameters, LaMDA uses external knowledge sources to improve the groundedness of its responses. Google collected a dataset of dialogs between people and LaMDA, annotated with information retrieval queries and retrieved results. They then fine-tuned LaMDA's generator and classifier on this dataset to learn how to call an external information retrieval system during its interactions with users.

Zero-shot Domain Adaptation

LaMDA's capabilities go beyond factual grounding. It can also perform zero-shot domain adaptation, as demonstrated by its ability to impersonate Mount Everest by simply setting its initial message to "Hi, I'm Mount Everest. What would you like me to know about me?" Everest LaMDA provided educational and factually correct responses.

Evaluation of Google LaMDA

Google AI conducted tests to quantify LaMDA's progress against key metrics. They collected responses from the pre-trained model, fine-tuned model, and human raters to multi-turn two-author dialogs, and then asked a different set of human raters to evaluate these responses against Quality, Safety, and Groundedness metrics. They found that LaMDA outperformed the pre-trained model in every dimension and across all model sizes. Quality metrics generally improved with the number of model parameters, with or without fine-tuning. Safety did not seem to benefit from model scaling alone, but it did improve with fine-tuning. Groundedness improved as model size increased, but fine-tuning allowed the model to access external knowledge sources effectively.

Future Research and Challenges

While LaMDA's level of Sensibleness, Specificity, and Interestingness presents new opportunities for understanding the benefits and risks of open-ended dialogue agents, there are still significant limitations. Google AI plans to continue exploring new ways to improve their Safety metric and LaMDA's groundedness, aligned with their AI Principles.

Availability for the Public

According to Google's Blog News, While LaMDA is still in development, Google AI plans to make it available to the public in the future. However, there is no official release date as of yet. Google AI is also committed to ensuring the ethical use of LaMDA and will be monitoring its deployment closely to prevent misuse.