Google is not used to play catch up in either artificial intelligence or search, but today the company is rushing to prove that it has not lost its edge. It is starting to roll out a chatbot called Bard to fight the sensational popularity ChatGPT.
Bard, like ChatGPT, will answer questions and discuss an almost endless range of topics with what sometimes seems like human understanding. Google showed WIRED several examples, including asking for activities for a kid who loves bowling and asking for 20 books to read this year.
Bard is like ChatGPT in that he will sometimes make up stories and act weird. Google has revealed an example where it misspelled the name of a suggested houseplant. “The Bard is an early experiment, it’s not perfect and sometimes it goes wrong,” said Eli Collins, Google’s vice president of research for Bard.
Google says it has made Bard available to a small number of testers. From today, anyone in the US and UK will be able to register access.
The bot will be accessible through its own website and separate from Google’s usual search interface. It will give three answers to each query—a design choice intended to impress the user that Bard is generating answers quickly and can sometimes make mistakes.
Google will also provide a suggested query for regular web search below each Bard response. And users can give feedback on its answer to help Google refine the bot by clicking the like or dislike icon, with the option to enter more detailed feedback.
Google says early users of Bard have found it a useful aid for generating ideas or text. Collins also acknowledged that some people have successfully misled it, though he did not specify how or exactly what restrictions Google has attempted to place on bots.
Bard and ChatGPT show great potential and flexibility but also unpredictable and still in the early stages of development. That presents a conundrum for companies hoping to gain an edge in promoting and exploiting technology. For a company like Google with large established products, this challenge is particularly daunting.
Both chatbots use powerful AI models to predict the words that will follow a certain sentence based on statistical patterns gathered from huge amounts of text training data. This turned out to be an extremely efficient way to mimic human responses to questions, but it meant that algorithms would sometimes invent or “hallucinate” facts—a serious problem when bots are supposed to help users find information or search. Webpage.
ChatGPT-style bots can also recall biases or language found in the dark corners of their training data, such as about race, gender, and age. They also tend to reflect the way users address them, making it easier for them to act as if they are emotional and prone to the urge to say weird and inappropriate things.
