Google unveiled the LaMDA 2 AI system at the Google I/O keynote this week. The new AI system is a step up from the original LaMDA which was designed for dialogue applications. The system could communicate with users and answer their questions.
The new LaMDA 2 is an enhanced version of the original which can engage in long, human-like conversations. Moreover, it can fetch accurate responses and mold them into easy-to-understand sentences.
How did LaMDA 2 come up with so accurate answers?
Sundar Pichai demonstrated the capabilities of the new dialogue system in his keynote speech. He showed a demo where a user asked LaMDA 2 about Mariana Trenches. It was a series of questions that the AI system had to answer one by one. Not only did it respond to queries but also answered the topics it wasn’t explicitly trained to do.
In another demo, the user asked about tips on planting a vegetable garden. The AI system touched on all the important topics related to plantations and offered tasks and subtasks to achieve the goal. Sundar Pichai’s demo of LaMDA 2 aptly displayed what we can expect from ML and AI in the coming decade.
Like all AI systems, LaMDA 2 was fed a lot of datasets to learn and train on how humans converse with each other. It is very rare for a dialogue to repeat which was an added layer of complexity. So, the AI system analyzed petabytes of data and trained to produce human-like answers.
Is it available yet?
LaMDA 2 will be available inside an app called AI Test Kitchen. The dates of availability are unknown as of now. Google plans to release it for researchers, academics, etc. It will help in improving the quality of answers. Correctness is still an issue and the Google CEO accepted the fact that not all answers would be correct. But it is a learning curve right now. Like most Google features, it will first roll out in the US.
Google recently unveiled another AI language model called PaLM which stands for Pathways Language Model. It applies a chain of thought processes to divide a reasoning problem into many steps. By tackling each step individually, the model produces correct answers. It can easily solve math problems and queries.
That’s not all from Google I/O. Google unveiled the new Pixel 6A and other products. What do you think of LaMDA 2? Will AI ever be capable enough to answer nuanced queries? Share your thoughts in the comments.