Google's new AI language model to take on GPT-4

Search engine giant Google has unveiled a new artificial intelligence. It goes by the name of "Gemini" and should be able to compete with OpenAI's GPT-4. Google CEO Sundar Pichai speaks of a "new era" in the press release. Gemini is a language model (Large Language Model or LLM for short) and will be the basis for various AI functions in Google's products.

The new model has three levels: Nano, Pro and Ultra. Nano is the smallest version, which requires the fewest resources. It is designed to run natively and offline on Android devices. Gemini Pro is more powerful and is now being used, for example, in the English version of Google's chatbot Bard. The highest level of the new AI, Ultra, is intended for professional applications and data centres and will not be available until next year.

The highlight of Gemini is that it is multimodal. This means it can handle not only text, but also images, audio and video. Without plug-ins, as is the case with the market-leading OpenAI GPT-4 model. Gemini should be at least as good. According to Google, it beats the competition in 30 out of 32 benchmarks. Even if the gap is small in most cases.

AI as a tutor

Google is proud of its performance in a very broad benchmark, among other things. The press release states: "With a score of 90 per cent, Gemini Ultra is the first model to outperform human experts in MMLU (Massive Multitask Language Understanding). This test uses a combination of 57 subjects such as maths, physics, history, law, medicine and ethics to test both world knowledge and problem-solving skills."

As a concrete application example, Google shows AI as a tutor, for example when parents themselves have too little knowledge of a topic. For example, Gemini can analyse a maths problem with a handwritten solution and explain specifically where an error has occurred in the calculation. Questions can also be asked afterwards.

How well Gemini performs in practice remains to be seen. In any case, Google seems determined not to lose out in the race with OpenAI. Just under a year ago, the search engine giant was caught off guard by the success of ChatGPT and subsequently declared a "code red" internally. Google then launched its own chatbot Bard head over heels, but it was never really able to keep up with ChatGPT.