In the AI field, OpenAI has successfully captured the world's attention. But not for long, at least that is what Google wants.
When ChatGPT introduced ChatGPT, the world was quickly captivated by the AI's prowess. It's nothing like what the world has ever seen before, and it's a lot powerful, and a lot smarter than any chatbots or the so-called digital assistants that came before it.
Google, which has merged its AI team with DeepMind, the company it acquired it 2014, claims that 'Gemini' is its next large language model that will rival ChatGPT.
In fact, if everything goes as planned, the so-called 'Gemini' should be able to have the ability to plan or solve problems as well as analyze text, said DeepMind CEO Demis Hassabis.
In other words, the AI from Google DeepMind, should have the ability to surpass ChatGPT.
To achieve that height, Google DeepMind is using its knowledge and techniques the team has acquired from the development of AlphaGo, DeepMind’s AI system that was the first to defeat a professional human player at the board game Go.
The Gemini AI was briefly teased at Google’s I/O developer conference in May 2023, in which CEO Sundar Pichai noted in a blog post that Gemini is the company’s "next-generation foundation model."
According to the company, the AI shall leverage innovations in reinforcement learning to accomplish tasks with which Large Language Models from rivals often struggle.
The method involves reinforcement learning by "rewarding" an AI system for certain behaviors, but punishing the AI system for certain undesired ones.
The goal is to teach the AI system which behaviors to exhibit in a given situation.
This method isn't exactly new, and has been widely used to make AI models understand what they have to do. In fact, reinforced learning is also part of ChatGPT's training.
But DeepMind has a lot more wealth of experience in reinforcement learning, and thanks to that, AlphaGo and some other powerful AIs were born.
And as for the sources of the training materials, it's reported that an anonymous source said that Google’s researchers have been using YouTube.
YouTube, being the largest video-streaming platform, is sitting on a wealth trove of data. And because it's Google's property Google DeepMind can tap into it to get "more complete access to the video data than rivals that scrape the videos."
With the experience, Google DeepMind, which is a merge between Google's AI team and DeepMind, is applying its learnings to the generative AI domain.
In other words, Google DeepMind is trying to beat OpenAI at the game Google DeepMind knows better.
Unlike its rival, which is powered by GPT-4, Google DeepMind's Gemini should be able to handle any data or any task without requiring specialized models. Once fully developed, Gemini should have the potential to offer more creative responses too, by generating ‘off-script' content that isn't restricted by its training data, and that's based on the structures it learned during training instead.
Read: Google's AI Beats Human Go Champion
"At a high level you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models," Hassabis said. "We also have some new innovations that are going to be pretty interesting."
It's worth noting that as a company with tons of resources and support from Google, Google DeepMind has lots of other projects.
But at this time, thanks to OpenAI that created the hype of generative AI and sent others scrambling, Google has showcased Bard.
And Gemini is Google DeepMind's most ambitious work in the space to dethrone ChatGPT.
It's said that Gemini was created when Bard became Google's $100 billion mistake.
Gemini is yet another answer from the tech giant, who just couldn't bear seeing its rival shine.
Gemini is a product Google’s DeepMind planned because the company couldn't watch OpenAI's success in silence and not doing anything about it.
It's a product Google is attempting to use to overcome ChatGPT from an already-losing position.