Google's DeepMind AI is quickly catching up with Microsoft

Google’s partnership with DeepMind, suggestively called Google DeepMind, is about to come out with the strongest AI model on the market, as per their accounts.

The AI model is called Gemini and in an interview for Wired, Demis Hassabis, DeepMind’s CEO, says that the new AI innovations coming from Google are very interesting, to say the least.

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 [e.g., GPT-4 and ChatGPT]. We also have some new innovations that are going to be pretty interesting.

Demis Hassabis

Gemini will be released later this year, in an effort to combat Microsoft’s lead in AI research. As you might know, Microsoft has been at the forefront of AI innovations. Models such as Orca 13B, phi-1, Kosmos-2, LongMeM, or the more recent CoDi, promise a new era for AI. And they’re all happening now.

Even more, it seems that Gemini is being touted as a breakthrough in AI technology. The model will greatly disrupt Microsoft’s OpenAI models on the market, specialists say.

Could Google DeepMind AI, Gemini, be a breakthrough?

Gemini is part of DeepMind’s Alpha family of AI models. Other models from this family are AlphaGo, AlphaGo Zero, AlphaZero, and MuZero. And they are trained in limited or constrained environments, so they are often pushed to the limits. This way, these models surpass human abilities and even human knowledge.

The training on Alpha models happens under human supervision, and the results improve when these models are left to their own devices to find solutions.

On the other hand, GPT models, such as Bing AI or ChatGPT, Orca 13B, and others ones, are trained on huge quantities of data. The training on GPT models does not require human supervision, as these models learn to train themselves.

If you remember, phi-1, and Kosmos-2 were models that learn to code and visualize space on their own.

Gemini would be a combination of those two types of training, and it would be the way to the AGI, and then ASI. In other words, Gemini would mean a step closer to Artificial General Intelligence, and then Artificial Super Intelligence.

We’ll have to wait until fall, or winter 2023 to see if Gemini holds up to the hype. But the time of AI is here. And it seems Google does not want to lose ground to Microsoft when it comes to AI. Innovation breeds competition.

What do you think about it? Let us know in the comments section below.