The service is now also starting in Germany.
(Photo: IMAGO/NurPhoto)
Dusseldorf Almost four months after the market launch, Google is now making its AI chatbot Bard available in Germany and throughout Europe. A total of 59 countries are added, as the group announced on Thursday. The delay was due to EU privacy concerns. “We had productive discussions with the authorities,” says Jack Krawczyk, Google’s product director.
The Internet giant played it safe in the past few months and initially only made Bard available to a small group of users in the USA and England. Japanese and Korean were added in May. The artificial intelligence (AI) now speaks more than 40 languages, including German, Spanish, Arabic and Chinese.
Google is under pressure: The chatbot ChatGPT has been available everywhere for many months. Google rival Microsoft has invested billions in its developer OpenAI and has already integrated the technology into its Bing search.
Google’s AI chatbot Bard can now also process photos
Experts now welcome Google’s move. “The EU is indispensable for the group simply because of its economic size,” says social media analyst Matt Navarro. “But the delay also shows how powerful the EU is with its rules and regulations.”
In principle, Google did not change the language model for Bard. The Palm 2 language model continues to be the basis for the text robot. However, it has been fine-tuned, explains Krawczyk. There are also new functions. Bard can now read out his answers, and you can also set the “tonality and style”, as Krawczyk says. For the time being, however, this is only in English.
More on artificial intelligence and ChatGPT:
An important innovation is the use of images: users can post photos to get information or “inspiration for a text,” says Krawczyk. The Google manager gives an example: He photographed the labels of wine bottles. He later asked Bard for suitable recipes: “I should use Pinot Noir for grilled chicken, for example.” This function is also only available in English for the time being.
The introduction of so-called “pinned threads” is interesting. These “pinned threads” allow users to save their dialogues with the AI. With the option to “pin, rename and go”, not only does information not get lost, but the quality of the conversation increases. Otherwise, language models “forget” dialogues and cannot give exact answers.
Users complain about “good” answers from Bard
Krawcyzk did not want to say how many questions are allowed on a topic and in a conversation. With persistent questions, various users have often led text robots to violate the rules, for example to make obscene or violent statements.
Google wants to prevent the so-called “jailbreaking” at all costs, out of concern for the image and trouble with advertisers. “We put a whole team in place to get Bard to make statements like this in order to exclude them,” says Krawczyk. There is also such a team in Germany.
Accordingly well-behaved, Bard answers questions. “The answers are a little boring,” says analyst Navarro, who has been comparing the AI to ChatGPT for a few weeks. “That’s why I always prefer ChatGPT.”
Google’s AI chatbot Bard: “I’m not a person”
Bard boss Krawczyk promotes understanding. “Some of Bard’s answers can frustrate users.” But it is more important to present a “responsible model” that is at the same time helpful and not harmful. Overall, Bard is used a lot. However, the manager did not want to name a number of inquiries or users. He spoke of a “high return rate” among users.
Katharina Zweig, Professor of Computer Science at the University of Kaiserslautern, has observed something “exciting”, as she puts it. The text robot does not answer with “I”: “It was explicitly trained out of this, so that the impression would not arise that an I is speaking with a consciousness.”
However, Bard cannot do without it entirely. For some inquiries from the Handelsblatt, the text robot answered in the first person singular. “I’m not a person in the traditional sense. I’m a powerful tool,” it replied when asked if it was a human. Then why does it use “I”? “It’s a way of communicating that’s familiar and understandable to people.”
More: What can Google’s AI do?
First publication: 07/13/2023, 9:06 am.