The US company Alphabet wants to outperform the competition from Open AI and Microsoft with the AI language model Gemini, which was published on Wednesday. The new AI system comes in three variants, of which only two are currently available. Gemini Pro, which despite the name can be used free of charge, can now be found as a basis in Google’s chatbot Bard and will gradually be integrated into all Google products. However, users from the EU will have to be patient for a while.
While prompts to Google’s chatbot are processed in Google’s data centers, Gemini Nano is intended to run locally on Android phones. However, this currently only works on a single phone model, Google’s Pixel 8 Pro, which has Google’s Tensor 3 chip installed. Gemini Nano is intended to improve more and more functions on the phones, but at the moment it is only being used in the auto-reply function on WhatsApp and for summarizing recordings with Google’s voice memo app Recorder.
Gemini Ultra, on the other hand, is said to be so good that it can compete with the latest version of Chat GPT – 4 Turbo. In tests Gemini Ultra performed better across the board than Open AI’s AI, although not much better. Before consumers can test the model themselves, they will have to wait until at least January; according to Google, final safety tests are still being carried out for the model.
Not in the EU, only from January, only on the latest Pixel – that’s quite a lot of restrictions for the Bohei that Google threw at the introduction of the new product. The company is apparently sure that its AI model is a great success despite these limitations. The magic word in this context is “multimodality”. Specifically, this means that Gemini can work with many different inputs. This includes text, audio, images, programming languages and video. Unlike previous AI models, Gemini did not combine image recognition and text recognition, but rather Gemini was trained directly with various inputs. This is intended to ensure that the AI can handle the various sources more intuitively and thus make more complex queries possible.
The painted guitar suddenly plays 80s rock
The interaction with images and videos appears particularly impressive in small example videos. For example, the model was presented with a physics homework assignment, which corrected the model and was able to directly explain where the student was in the text assignment which way he took a wrong turn. In addition, Gemini can directly generate new word problems to test whether the student would make the mistake again. Physics and mathematics teachers all over the world could save themselves a lot of work this way, but they are also not entirely afraid of losing their jobs.
This AI ability is not entirely new. Chat-GPT 4.0 can also easily correct the physics task given to Gemini. What is new, however, is that Google already wants to offer this service in the free version in Bard. If you want to use Chat-GPT 4.0 including image analysis and generation, you currently have to pay $20 a month.
Gemini can apparently also react to live video. In a short clip, a Google employee paints a guitar, which Gemini translates directly into music. When an amplifier is added, an electric guitar immediately sounds. A little later, the employee plays rock, paper, scissors with the AI. This may be an inconsequential gimmick, but it is a step forward compared to previously available platforms.
Scientists could also take a look at Gemini. The model can search through countless scientific articles for passages relevant to researchers, supplement old articles with new data, even expand old tables with new data. Work that would have previously taken assistants days can theoretically be completed in just a few minutes.
Of course, as always with grandiose announcements, caution is advised. Although many US YouTubers have made their own attempts with Gemini in the past 24 hours and are largely full of praise for the new possibilities, most of the impressive demos so far have been from Google itself. They were able to convince themselves of Gemini’s capabilities SZ not yet. Despite this limitation, one thing is comparatively certain: after Google’s botched AI launch with the error-prone chatbot Bard, the company has created a real alternative to Chat-GPT with its current model.
The stock market also reacted positively. The price is currently over two percent higher than before Google’s announcement. It is still unclear how much chatbot AI threatens Google’s most successful business model – advertising in web searches.