The technology giant Google has introduced its most capable artificial intelligence model so far. With that, the company’s Bard chatbot has also received a significant update – although only outside of Europe.
Google’s Gemini artificial intelligence model is becoming part of the company’s most important services. rafapress
The Gemini artificial intelligence model launched by Google on Wednesday is one of the technology giant’s biggest science and engineering projects of all time.
It’s part of a huge upheaval that the CEO of Google’s parent company, Alphabet Sundar Pichai even calls the introduction of the internet more significant.
– Artificial intelligence will promote knowledge, learning, creativity and productivity on an unprecedented scale, states Pichai In a Google blog post emphasizing that humanity has only scratched the surface when it comes to mapping the possibilities of artificial intelligence.
Not to Europe yet
Gemini is a so-called multimodal artificial intelligence model, meaning it is able to generalize, understand and seamlessly combine different information, such as text, image, sound, video and different coding languages.
Until now, multimodal models have struggled with more conceptual and complex reasoning.
Gemini 1.0 was launched on Wednesday in 170 countries. In Europe, however, Bard users still have to hold their breath while Google waits for the green light from European Union legislators.
Google’s Bard competes with OpenAI’s ChatGPT and Microsoft’s Bing chatbot, among others.
Three versions
There are three different versions of Gemini in total; Gemini Pro, which is integrated as part of Google’s Bard chatbot, Gemini Nano arriving on Android phones, and the top version Gemini Ultra, which is still being worked on, will be baked into the more advanced and soon-to-be-released Bard Advanced chatbot.
Gemini will eventually become part of Google’s search engine and the Chrome browser.
Although it is not yet available for consumers to try, information has already been received about Gemini Ultra’s capabilities. Its skills were tested in the MMLU exam, which combines questions from no less than 57 subject areas, such as mathematics, physics, history, law and medicine, and ethics.
Gemini Ultra scored 90.0 percent on the MMLU test, which is said to be better than human experts.
According to Google, Gemini Ultra considers and weighs the questions presented to it more thoroughly than other AI models, which improves its problem-solving ability in particular. However, Gemini Ultra scored worse than the GPT-4 model for reasoning tasks requiring common sense.
The problem is “hallucinations”
Although Gemini can be, so to speak, better than humans in some tasks on a good day, the answers it gives should be treated with great caution.
Gemini’s weakness, like other generative artificial intelligence models, is the surprising “hallucinations”, i.e. incorrect and completely out-of-the-box answers presented as truths.
Sources: Google, The Guardian