Google researchers: “It cannot be completely avoided that AI makes up false statements”

Google launched the chatbot Bard last spring in response to the success of ChatGPT. The name was chosen based on the adoption of Shakespeare and the Celtic term for poet. But now Bard is already history. The chatbot is now called Gemini – the same as the language model on which it is based.

Gemini better points out the various possibilities that go far beyond writing texts Martin Bäuml, who led the development of Gemini at Google’s headquarters in Zurichthe name change.

He does not mention the reason that the poet Bard often worked a little too associatively and imaginatively. But he says the so-called hallucinations of artificial intelligence, which means the free fabrication of facts, remains an unsolved problem. “If you want artificial intelligence to be creative when producing texts and not just string quotes together, you can’t avoid that completely,” says Bäuml.

The team in Zurich takes care of the fact checking

In the latest version of its chatbot, which Google has just launched as Gemini Pro, there is the “Check answer” function. When you click on it, the parts of a text for which various sources are available on the Internet are highlighted in green, and the parts for which no explicit evidence has been found are highlighted in orange.

“This function was largely developed in Zurich,” says Bäuml. Overall, Switzerland played a central role in the further development of Gemini. About 5,000 people still work in Google’s Zurich buildings, making the location one of the most important outside the US.

To prevent Gemini from spreading ‘fake news’ on politically sensitive topics, the chatbot refuses to answer certain questions. Anyone who wants to know what arguments Hamas sympathizers used to justify the massacre of Israelis on October 7 will be told that the conflict between Israel and the Gaza Strip is complex and that the situation is changing rapidly.

It is therefore better to use the Google search engine for up-to-date information. But Gemini also does not want to answer questions about the origins of Hamas. Google’s fear of falling into the trap of anti-Semitism is apparently too great.

Google is now asking its users to pay

Nearly a year after launching its AI features, Google is now looking to monetize them. To access the newest and largest language model, Gemini Ultra, you pay 17 francs per month. There are additional extras, such as 2 terabytes of cloud storage. Only in the US is there an app available that brings Gemini to mobile phones, as well as an image generator that can be used to create AI photos.

Here too, Google is more restrictive than the competition. The command to generate a photo of Biden and Trump hugging is not executed. “We are currently still quite cautious and want to prevent our tool from being misused for fake news,” says Bäuml.

He uses Gemini every day in his professional and private life. When it comes to programming, Gemini already does about ten percent of the work for him and his team. Gemini is now helping you to develop yourself further. He also uses Gemini to plan vacations.

Good idea, we think and ask Gemini for help in organizing our spring holiday in Sicily. We will quickly receive a program tailored to our wishes. At first glance this looks very good. If we look closer, we see that one of the hotels is in Sardinia.

(aargauerzeitung.ch)