Google Gemini: Could This Be the Next Breakthrough in AI

Google’s much-anticipated next-generation generative AI, Google Gemini, has finally debuted. However, the current release of the AI is a scaled-down version of the more robust Gemini model scheduled for launch next year. Let’s backtrack a bit.

In an online press conference yesterday, Google’s team DeepMind, in collaboration with Google Research, provided a brief yet insightful summary of Gemini and its impressive capabilities.

What does Google Gemini offer?

Surprisingly, Gemini isn’t a solo act but a whole home of AI standards with three distinct versions:

1. Gemini Ultra is the top-tier flagship model.

2. Its pro model is a lighter version of Gemini.

3. Gemini Nano is crafted explicitly for mobile devices.

It’s a diverse lineup catering to various needs and device capabilities. Testing out the Pro version of Gemini is a breeze with Bard, Google’s rival to ChatGPT. Today, Bard is energized by a refined iteration of Google Gemini Pro, which only applies to English text in the United States (images are not included). According to the GM of Bard & Google Assistant Sissie Hsiao, the Gemini Pro brings enhanced planning, reasoning, and adapting capabilities compared to the earlier model powering Bard.

Gemini Inherently Multimodal.

Gemini Pro is the refined version powering Bard, but it doesn’t quite steal the spotlight. Hsiao says it excels in brainstorming, content summarization, and writing. Moreover, it outperforms GPT-3.5 and GPT-4 in six models,. It includes one (GSM8K) that evaluates grade academy math logic. However, this achievement could be more groundbreaking, considering GPT-3.5 is around one year old.

Gemini Ultra is also designed to be “Inherently multimodal” like Gemini Pro. In simpler terms, it undergoes pre-training and fine-tuning in various languages, images, codebases, audio, and videos. DeepMind Product VP ELI Collins asserts that the Ultra version of Gemini can grasp “nuanced” information across text, audio, images, and code, excelling in answering questions related to complex topics, especially in physics and math.

Recommended:

What do you all need to know about Google RankBrain?

What are the most Accurate AI detection tools?