Google has a new Artificial Intelligence model. The technology giant is going all out in the field of AI and is betting very heavily on the Gemini brand. So much so that it will leave GPT-4 far behind in a short time.
It’s only been just two months since the launch of Gemini, the large language model (LLM) that Google hopes will take it to the top of the AI ​​industry. However, despite the short time that has passed, the technology giant has presented its successor to go after OpenAI and GPT-4 once and for all.
The raw potential of Gemini 1.5
Broadly speaking, Google’s new version of Gemini can handle much larger amounts of data. The updated AI model can now do really impressive things with videos or long texts.
In the video of its presentation we have seen an example that serves perfectly to exemplify that the era of a chatbot that simply responds is past. The Google team uploaded a 44-minute silent movie starring Buster Keaton and asked the AI ​​to identify what information was on a piece of paper that, at some point in the film, is pulled out of a character’s pocket. In less than a minute, Gemini 1.5 found the scene and correctly identified the text written on the paper. The researchers also asked the model to find a scene from the movie based on a drawing, which it successfully completed.
Leaving ChatGPT behind
To put the potential of an Artificial Intelligence model into context, we have to take into account the concept of contextual window and tokens. This last term, somewhat complex to define, would be the unit of measurement of the basic components necessary to process information. As Google explains, “they can be entire parts or subsections of words, images, videos, audio or code. “The larger a model’s context window, the more information it can absorb and process in a given message, making its output more consistent, relevant and useful.”
Gemini 1.5 Pro has a huge context window, meaning it can handle much larger queries and see much more information at once. That window is 1 million tokens, a fairly considerable figure compared to the 128,000 of OpenAI’s GPT-4 and the 32,000 of the current Gemini Pro. Indeed, this is the largest context window ever seen in a large language model.
Of course, everything has a certain trick. For now, Gemini 1.5 Pro – and its million tokens – will only be available to enterprise users and developers, through Google’s Vertex AI and AI Studio. It will eventually replace Gemini 1.0 and the standard version of Gemini Pro, which is available to everyone at gemini.google.com and in the company’s applications, will be 1.5 Pro with a contextual window of 128,000 tokens, surpassing the current 32,000 tokens.
You will have to pay more to reach one million tokens. Google is also testing the ethical and security limits of the model, particularly as it relates to the new larger context window. When that happens and Gemini 1.5 with 128,000 tokens is the model that you can try for free on the project’s website, it will mean that the free version of the generative AI assistant will be comparable to the current paid version of ChatGPT Plus, which is the one used the GPT-4 model (or you can try it for free with Bing).
