Amid the hype around ChatGPT, Google announced the start of open testing of the Bard neural network. This is a generative chatbot that can respond to different user requests.
Izvestia found out how the experimental bot will work and how it differs from ChatGPT.
Product from Google
For now, Bard is only available to users in the US and the UK, but over time it will open to residents of other countries, Wired writes. When this will happen, the company has not yet disclosed, but they note that now the chatbot is working in experimental mode in English and can give inaccurate information.
Like OpenAI’s ChatGPT, Bard invites people to ask questions on various topics. At the same time, Google notes that the neural network will not replace the search engine, but will simply complement it and draw information from there.
The developers claim that Bard differs from ChatGPT by having more up-to-date information: ChatGPT uses data collected before 2021 in its responses, and Google’s neural network can replenish its knowledge in real time.
In addition, judging by the reviews, it works faster and gives more detailed information. But if a person does not like the answer, he can put a dislike so that Google engineers understand which parameters of the neural network should be improved.
(Not)justified risk
The Verge journalists who tested the new chatbot said that he was able to quickly answer a number of simple questions, for example, how to get a child to go bowling (“just take him there”) or make a list of films about robberies.
However, the actual information (for example, about the capacity of a particular washing machine) it is difficult to wait for a neural network. Although under each answer she offered a “Google It” button with a link to the corresponding query in the search engine.
Users believe that it will be possible to really check the chatbot only after full access is opened. When people around the world start accessing the neural network, its capabilities and disadvantages will become more understandable.
According to Wired, Google created its first chatbot back in 2020, but the management did not implement it, fearing risks due to inaccuracies in the neural network. However, amid the hype around ChatGPT from OpenAI, which in addition recently introduced the GPT-4 model, capable of analyzing pictures, writing complex texts and understanding 28 languages, the company decided to take a risk.
According to the interlocutors of the portal, Google ventured to launch a bot “far from perfect”, based on the considerations that the neural network will develop precisely through interaction with people. The company hopes that due to Bard’s mistakes, its reputation will not suffer.