Since the introduction of GPT 3.5, progress in the field of artificial intelligence has increased tenfold. Every week, new applications and algorithms are released, and major companies are racing to improve performance.
Having announced GPT 4 and the forthcoming release of GPT 5, Microsoft partner OpenAI seems to be taking the lead. Against this backdrop, Google DeepMind unveils their new algorithm, Gemini.
So, what is Gemini capable of? Does it have the potential to replace GPT? Let’s find out in this article about the potential of Alphabet’s new algorithm.
What is Google Gemini?
Unveiled at the Google I/O 2023 conference by Google CEO Sundar Pichai, Google Gemini is the first project of Google DeepMind, the union between Google Brain and DeepMind, two of Google’s artificial intelligence development sections.
With Google Brain being the founder of AlphaGo, an that beat Go world champion Ke Jie in 2017, and DeepMind a pioneer of reinforcement learning models, Google believes that new AI capabilities will be discovered through this merger.
Currently under development by the Google DeepMind teams, the final project should see the light of day in several months’ time, i.e. in 2024. What we do know, however, is that Gemini will be a generation, planning and problem-solving algorithm, like ChatGPT 4.
What are its features?
According to an interview with Demis Hassabis, CEO of Google DeepMind, Gemini will use technologies developed for and with AlphaGo and reinforcement learning techniques coupled with the linguistic capabilities of Google’s LLM, PaLM 2.
Gemini will be a multimodal algorithm, meaning it will be able to draw information from different data sources simultaneously. These users will then be able to provide it with video, graphical or textual data, so that it can develop more concrete answers and solutions.
While Gemini is not the first multimodal algorithm to be created, nor is GPT 4 or LLaMA, Google is approaching the use of this data in a way that sets it apart from its competitors. For Google, its technologies must be accessible to everyone, without compromising their privacy. That’s why the company has committed to using only royalty-free or public databases to drive its new model. The aim is to give AI an ethical image and counter the criticism levelled at some of its competitors.
Will Google Gemini replace GPT?
According to available information, Gemini should exceed GPT 4’s thousand billion parameters. A figure to be reconsidered when we know that LLaMA and its 70 billion parameters offer similar performance to GPT 4 in certain respects.
In terms of budget, Mr. Hassabis estimates that Gemini could cost several tens, if not hundreds, of millions of dollars, a figure close to the $100 million announced by OpenAI CEO Sam Altman for the creation of GPT 4.
Google DeepMind is under pressure to deliver a product that raises the bar for AI. If you’d like to learn how to master Gemini or any other AI tool to perfection, feel free to learn more about our AI engineer training or watch our blog posts to stay informed.
