How Google Plans to dethrone ChatGPT with Gemini, its upcoming family of large language models.
OpenAI’s large language models so far have been leading the AI race, thanks to their early launch and Microsoft’s vast data centre infrastructure to back them up. But the ChatGPT dominance may not last forever, as new and powerful AI models are emerging every month. And one of them has a far stronger chance of becoming a real challenger: Google.
According to a report by The Information, Google’s next-generation artificial intelligence models under the new Gemini project will launch as early as this fall. The company for Gemini to power its existing AI chatbot Bard and enterprise apps such as Google Docs and Slides.
What makes Gemini a formidable competitor is the amount of resources Google has at its disposal, especially data that can be used to train these AI models. Google has access to YouTube videos, Google books, a huge search index, and scholarly material from Google Scholar. Much of this data is exclusive to Google, and that could give it an edge in creating smarter models than others. Gemini is also said to be the first multi-modal model that can handle video as well as text and images, unlike GPT-4.
Google also has a deep pool of talent and years of experience in building and training large language models. It is expected to announce the new models next fall, and possibly introduce a new Gemini-powered chatbot or upgrade its existing Bard chatbot. This could have a big impact on Google’s Cloud, which would likely be the main way for corporate customers to access Gemini.
Gemini was first teased at Google’s developer conference last month, where Google showcased several AI projects. Gemini uses new training techniques from Alphago, an AI system that was the first to beat a professional human player at the complex board game Go, according to a report from June. This could enable Gemini to plan and solve problems.
It’s a team effort
In April, Google merged Google Brain, its own deep learning artificial intelligence research unit, with subsidiary company DeepMind to create a single unit – Google DeepMind. The surprise move by Pichai was meant to boost efficiency, combining Google’s enormous computational resources and DeepMind’s research skills.
Before that, both entities had their own responses to ChatGPT. DeepMind had Project Goodall and Google had Bard based on Google Brain models. Despite some rivalry between the teams, DeepMind dropped Goodall to collaborate on Gemini.
The end of ChatGPT’s reign?
The fact that Google Brain and DeepMind are working together on this could spell trouble for OpenAI and other rivals. How Google is building Gemini also matters. Reports suggest that Gemini has made significant progress in its multimodal abilities, surpassing previous models. Its design focuses on being multimodal, meaning it can process and understand different forms of data, as well as being highly efficient in terms of tools and API integrations.
This means that Gemini will not only be good at understanding and generating conversational text, but also proficient in dealing with various inputs, such as text, images, and videos. There are also reports indicating that Gemini is being trained on twice as many tokens as GPT-4 was, which should make it much smarter.
Source:indianexpress.com