Google’s Gemini will appear in various guises and is expected to outperform GPT-4. In this way, Google wants to show that it and no other is the AI ​​company par excellence.

Sundar Pichai (Google CEO) and Demis Hassabis (Deepmind CEO) proudly introduce Gemini. Gemini is the new cutting-edge LLM to emerge from Deepmind’s laboratories. Just a few days ago it looked like Gemini wouldn’t see the light of day until 2024, but Google couldn’t wait to make the announcement. With Gemini, Google wants to once again show the world that it is an “AI-first” company, a reputation that has diminished somewhat since the launch of ChatGPT. Twins appear in three forms: Ultra, professional And Nano.

Google doesn’t shy away from comparison with GPT-4. One blog lists 32 benchmarks, of which Gemini Ultra performs better than the competition on thirty. Gemini Ultra is also said to have achieved a score of ninety percent in multitasking tests that combine different topics. This makes Gemini the first AI model to perform better in these tests than human experts. The AI ​​model is also capable of more complex reasoning.

To train Gemini, Google has also developed a new generation of TPU for its data centers: Cloud TPU v5p. Despite the expanded capabilities, applications based on Gemini models could run faster than less “heavy” counterparts.

Big upgrade for Bard

The theory sounds great, but what about in practice? As a mere mortal, you have the best chance of coming into contact with Gemini Pro. Google wants to widely deploy the mid-range version of Gemini in its own ecosystem, starting with Bard, which is receiving its biggest update yet. Google launched Bard in February to compete with ChatGPT, but the chatbot never achieved that level of success. That has to change with twins.

Thanks to Gemini, the chatbot has better reasoning skills, which should lead to better answers that get to the point. But Bard will also expand his programming knowledge and add multimodal skills. The latter means that Bard is not limited to text-to-text, but can also process and produce image, audio and video.

Through its sister channel YouTube, Google is showing a demo of how a well-known influencer builds a paper airplane with the help of Bard. We would like to point out that such functionalities are also included in the latest version of ChatGPT. This makes it even easier to stay up to date with what Google is doing. Bard’s new features are currently only available in English, currently in 170 countries.

Google plans to release a “Bard Advanced” version that runs on Gemini Ultra early next year. Gemini Pro will gradually be rolled out to the search engine, Chrome and other Google services. The model is also offered to developers in Google Cloud via Vertex AI.

AI on your smartphone

To ensure you have the best AI has to offer today, anytime, anywhere, Google launches Gemini Nano. This is a “smaller” version of the LLM that was developed specifically for use in smartphone processors. Google has already prepared its latest Pixel smartphones for this by incorporating a TPU into the Tensor chip. This means that many new AI functions, including image processing, will be rolled out to the latest devices with software updates.

The most powerful version, Gemini Ultra, will remain in the waiting room for a while. Only a limited number of customers are permitted to experiment with the model in a controlled environment. Gemini Ultra takes a big step towards general artificial intelligence, where AI can fully compete with humans. Therefore, Google wants to conduct even more extensive security testing before the technology gets into the hands of the general public.

In any case, with the launch of Gemini, Google is showing that it is once again fully participating in the AI ​​race. It is clear that the hype around generative AI will not die down immediately in 2024. While the focus this year was primarily on playing and experimenting, the added value of the technology will probably only become fully visible next year.