As of now, researchers estimate the generative AI market to be worth $1.3 trillion by 2032, and ChatGPT from OpenAI has currently reached more than 100 million active monthly users. Both these might have been what prompted Google to take the necessary steps to soon inform their presence in the generative AI market, and the company did so with its Bard, a conversational AI chatbot. The company has also unveiled a new language model (PaLM 2 LLM) in recent months.
Google has now introduced its GPT-4 direct competitor, Gemini, with 5 times more computing power and also with access to the vast array of proprietary training data that Google has. It is said that Google plans to release the latest language model to the general public by the end of this year, in the month of December 2023.
Here’s more about what we know.
Google Gemini
Speaking about the AI model, Google’s Gemini is trained on the brand’s TPUv5 chips, with the capability to simultaneously operate with 16,384 chips, and the dataset that has been used to train this model comprises around 65 trillion tokens. Contents from YouTube as well as training techniques that are inspired/taken from AlphaGo are also part of its training.
There are three main reasons that are being pointed out that will let Google’s Gemini outperform OpenAI’s GPT-4. The first is the AI model’s ability to generate texts and images. The second and most threatening reason is the proprietary training data that Gemini has from various Google services. And the third, is the partnership between Google’s DeepMind and Brain teams, bringing together Sergey Brin(Google Co-Founder), Paul Barham(DeepMind senior AI scientist and machine learning expert), and other AI researchers, suggesting that more can be expected.
We’ll have to wait to see how things will turn out and also to know whether Google’s Gemini will take the crown away from GPT-4 or not.
