The search giant has now opened up the model for public testing, and has again warned people to be cautious about what they say to it. “We’ve taken steps to mitigate risks, but there is always a potential for inappropriate content to appear,” Google said.
LaMDA is based on a natural language understanding (NLU) system that was trained on more than 12 billion words of dialogue to enable it to understand questions. The chatbot will then generate responses in natural language. What sets LaMDA apart from other chatbots is that it can also generate its own questions based on the answers it receives. This means it can generate conversations without any pre-scripted dialogue.
“In addition to generating its own questions, LaMDA can answer questions about arbitrary topics, even if it has never encountered those topics before,” Google said. “This allows for truly open-ended conversations.”
The chatbot is designed for entertainment purposes and it does not have any commercial applications at this stage. Google said it will use this public testing period to “gather feedback about the model’s capabilities and safety”.
The search giant has also released an API for LaMDA, which will allow developers to experiment with the chatbot.
More posts
