Google today announced lamda 2 on I / O 2022, a follow-up to the company's artificial intelligence system launched on Google I / O in 2021 Lamda 2 is the abbreviation of "dialogue application language model". It can decompose complex topics into simple and understandable explanations and steps, and generate suggestions when answering questions.
Lamda 2 is an artificial intelligence system established for "dialogue application". It can understand millions of topics and produce "natural dialogue", which will never take the same route twice. Like most AI systems, lamda 2 learns the possibility of words appearing in a text body, usually a sentence, based on many text examples. Examples appear in the form of files in training data sets, which contain TB to Pb data scraped from software hosting platforms such as social media, Wikipedia, books, GitHub and other sources on the public network.
On stage, Google CEO Sundar Pichai watched a demonstration in which a user asked lamda 2 to describe the Mariana Trench with a series of questions. The model answers questions about what creatures may live in the trench, as well as other questions about topics it has not been explicitly trained on, such as submarines and bioluminescence. In another demonstration, lamda 2 provides tips on planting a garden, a list of tasks and subtasks related to the location of the garden, and things that may be grown in the garden, such as tomatoes, lettuce or garlic.
However, considering that the predecessor of the system sometimes provides untrue "facts" in internal tests - repeatedly providing false information about Mount Everest in one case, the jury has not reached a conclusion on the cross task accuracy of lamda 2. Pichai acknowledges that models like lamda 2 are not perfect, but he highlights the complexity of the technology's higher-level capabilities and promises that Google is working to address these shortcomings.
"These experiences show the potential of language models to help us plan and understand the world one day," Pichai said.
In addition to lamda 2, Google also released AI test kitchen, an AI demonstration center driven by models such as lamda 2. As an application, users can interact with the model in limited ways, such as exploring a specific theme (such as dog) and delving into sub themes (dog smell). Google said it would continue to add "other emerging areas of AI" to the AI testing kitchen, including natural language processing and other areas.
AI test kitchen will be launched in the United States in the coming months, but will not be made public. Google has not yet fully decided how it will provide access, and the company is weighing a helping hand to selected academics, researchers and policymakers.