SOPA Images/Getty Images
- Google has revealed its GPT-4 rival Gemini, with users able to access an early version through Bard.
- But some have pointed out that the chatbot still can't get basic facts right despite its upgrade.
- Bard also seems reluctant to discuss controversial topics, telling people to Google it instead.
Google has finally unveiled its GPT-4 rival Gemini — but it doesn't seem to be any closer to solving its AI hallucination problem.
Social media users have been calling out Google's Bard chatbot, which is now running off a limited version of its GPT-4 rival Gemini, for continuing to "hallucinate" incorrect answers to simple questions — and avoiding controversial topics by telling them to Google it instead.
They pointed out that the chatbot didn't seem to know who won best actor at the 2023 Oscars, and struggled to understand requests to provide words in other languages.
Techcrunch initially reported the issues with Bard.
Users on X also noticed that when asked to provide updates about topics that could be controversial, such as Israel's ongoing invasion of Gaza, the chatbot declines to answer.
Instead, it tells users to "try using Google search" to get "up to date information." That's in contrast to rival chatbots that also have access to real-time information, like Elon Musk's Grok, Bing, and ChatGPT Pro, which give more detailed answers.
Google has always been clear that Bard, like other generative AI chatbots, can be inaccurate, and warns Bard users that it may "display inaccurate info, including about people, so double-check its responses."
It unveiled Gemini on Wednesday after months of anticipation. The AI model is seen as a huge moment for the company as it tries to catch up with rivals such as OpenAI.
Google was famously caught off guard by the explosive popularity of OpenAI's ChatGPT chatbot when it launched last year.
The tech giant released its Bard chatbot in February as it scrambled to respond, but faced embarrassment when ads for the ChatGPT-rival showed it giving an incorrect answer.
Google is rolling out Gemini in stages, with Bard currently running off "Gemini Pro." The company is planning to roll out the most advanced version of the model, known as "Ultra", early next year.
Ultra apparently outperforms OpenAI's GPT-4 model, according to Google, but the Pro version doesn't quite meet that benchmark, with the company saying that it performed better than the GPT-3.5 model that powers the free version of ChatGPT.
Google has already faced questions over the capabilities of its newest AI. A slick demonstration of Gemini responding fluently to physical objects and complex voice commands wowed the internet on Thursday.
However, a Google representative later told Bloomberg that the demo reel had been edited to make it seem more impressive, a discrepancy that was reportedly called out by some of the company's employees internally.
Google did not immediately respond to a request for comment from Business Insider, made outside normal working hours.