Google announced during Google I/O 2022 this yearAI Test Kitchen Project, allowing everyone to continuously train the natural language model through online interaction, so that the natural language model can respond more naturally and accurately in interactions. The trial is now open to everyone. Initially, it will only be available on Android platform devices, but the iOS platform will also be opened for testing in the coming weeks.
However, currently only a small number of users in the United States are eligible for the test, and Google will conduct a lottery internally to determine whether they are eligible for the test.
The AI Test Kitchen project uses Google's LaMDA natural language model and, through a project called "Imagine It," allows users to say an imaginary or real place, and the system attempts to describe it in words. In another "List It" project, users will provide shareable goals and thematic content, and the system will attempt to automatically categorize the content.
In addition, LaMDA can also have free conversations with users on specific topics in the "Talk About It" project, but currently it is set to let LaMDA focus on topics related to Gou, and will also avoid going off topic during the conversation with users.
In order to prevent users from maliciously interfering with training results, Google said it has implemented multiple layers of protection for the AI Test Kitchen project, such as automatically detecting and filtering all kinds of hateful, pornographic, violent or more illegal content that violates regulations. It will also prevent user privacy leaks and reduce the chances of the LaMDA natural language model being "broken" by others.
But will this happen again? Some people believe that the LaMDA natural language modelSituations with "human nature"Google said that there are still potential concerns, but stressed that it will minimize the related risks.
