Google’s new language model, Gemini, could soon come out and do just as well as the dominant one in this sector, GPT-4.
An AI, or more precisely, a chatbot, is an interface built from a language model, which gives it its power. For example, ChatGPT works with GPT-4, while Bard is based on PaLM 2. Unfortunately for Google, the latter, although efficient, was not yet up to the OpenAI language model. Reason why the giant has developed another project, which should soon be made available.
After PaLM 2, Gemini?
Gemini, this is the name that could well be on everyone’s lips in the weeks to come. This language model is indeed proof that Google is far from having given up in the face of OpenAI’s rapid advances in artificial intelligence. This should be much more powerful than PaLM 2, according to the first information obtained by The Information.
Gemini should in fact operate chatbots as much as serve as an assistant to its user, allowing them, for example, to write draft emails or obtain the latest news. And above all, it should join GPT-4 on a very important point.
Multimodal, the big breakthrough
Indeed, unlike PaLM-2, Gemini is a multimodal language model, which means that it will directly generate images or code from a prompt typed by the user. In this sense, it would even be a little ahead of GPT-4, another multimodal language which recognizes images when they are submitted to it without being able to generate them (even if DALL-E could quickly arrive on ChatGPT, according to the latest rumors).
For the moment, access to one of the latest versions of Gemini has been opened to a panel of developers, although, according to The Information, it is not the best version of the model. This would be, still according to the same source, at the level of GPT-4. A hell of a promise, which we are now impatiently waiting to see come true!
Sources: Reuters, The Information, Android Police
2