Prabhakar Raghavan, Google's vice president of search, sent an email to Google employees:
In the letter, Raghavan notes that with the helpBard internal testing, employees will help speed up the training of the model and test its resistance to load. Google employees are encouraged to ask the AI questions about topics they are knowledgeable about.
Recommendations attached to the lettertell you what you can and cannot do when training a chatbot. For example, responses should be polite, casual, and accessible. At the same time, they should be written in the first person and using an impartial and neutral tone. In addition, you need to avoid "animating" AI and hints that it has a "human-like experience".
This clarification is probably related to the previous onescandal when one of the Google employees said that the LaMDA language model has consciousness and a mind of its own. This claim was denied by company officials.
Bard - Google development, publictesting of which was announced last week. According to Google itself, the LaMDA model is used in Bard's work. According to the developers, this AI should compete with ChatGPT, which was launched earlier by Microsoft and aroused great interest in society.
Bard's first public presentation didn't go wellsuccessfully. The company launched a Twitter commercial showing a chatbot answering questions. Users noticed that the answers contain factual errors. As a result, the company's shares on the stock exchange fell by almost 10%.
Bard AI commercial with an erroneous answer: a neural network erroneously credits the James Webb telescope with the first image of an exoplanet located outside the solar system. Video: Google
Previously, Hi-Tech talked about the "arms race" that began between Google and Microsoft.
Read more:
It's not about the Earth: scientists explained why the solar system is the rarest
Dinosaurs in bamboo corsets were found among the artifacts of the expedition of the early twentieth century
Substance from the "grandfather's beard" improves memory and neuronal growth