Meet Google Gemini – The AI that Seeks to Dethrone GPT-4

Unveiled yesterday, Gemini is Google's most capable and versatile AI model to date– set to give OpenAi’s GPT-4 a run for its money. 

This natively multimodal family of AI models includes three versions, from basic to advanced, and can process text, images, videos, audio, and code to assist in multiple ways. 

Google Gemini is already integrated into Google Bard and Pixel 8, with more advanced applications coming up soon. 

This way, Google hopes to surpass GPT-4 regarding multimodal AI capabilities and usefulness and help developers build more advanced AI-powered tools at scale. 

Google Gemini is a Natively Multimodal AI Model

According to the formal introduction of Google Gemini, the new development is the most capable and flexible AI model they have produced to date. It is a groundbreaking achievement resulting from collaborative efforts across various Google teams. 

It is also one of the first natively multimodal AI models: instead of having each mode (around data format types such as images or text) trained separately and later stitched together, Gemini has been trained on different modalities from the get-go and later fine-tuned for added effectiveness. As a result, it can  “seamlessly understand and reason about all kinds of inputs (…), far better than existing multimodal models” and is said to be able to handle highly complex tax much better than its number one competitor, the widely praised GPT4 by OpenAI. 

Right now, the first version of Google Gemini splits into three different size models: 

• Gemini Ultra is the largest and most capable model for highly complex tasks.
• Gemini Pro is a standard model and the suggested one for scaling across a wide range of tasks.
• Gemini Nano is the basic bit intended for on-device tasks.

Google Gemini’s Outstanding Performance

Google doesn’t spare words when it comes to highlighting Gemini’s exceptional performance and, especially, how it outperforms similar models –they’re coming for GPT4, no doubt. 

They mainly remark on the model’s exceptional performance on various benchmarks, which are standardized tests used to evaluate the capabilities of AI models. For instance, Gemini Ultra achieved a remarkable score of 90.0% on the Massive Multitask Language Understanding (MMLU) benchmark, surpassing human experts. It also excelled in other benchmarks related to text understanding, coding, and multimodal tasks.

Another notable point is its advanced coding abilities, understanding, explaining, and generating high-quality code in popular programming languages like Python, Java, C++, and Go. Its proficiency in coding tasks, as demonstrated in benchmarks such as HumanEval and Natural2Code, positions it as a leading model for coding applications.

Last but not least, Google strongly emphasizes responsible AI development and safety. Gemini undergoes comprehensive safety evaluations, including assessments for bias and toxicity. The model is subjected to adversarial testing techniques to identify potential safety issues in advance. Google is also actively engaging with external experts to ensure a thorough evaluation of the model's performance across various topics.

Gemini in Google Products: Bard, Pixel and More

Since yesterday, several Google products have integrated Gemini capabilities into its features. 

Google Bard, the company’s conversational chatbot, is now fueled by a fine-tuned version of Gemini Pro, enabling enhanced understanding and reasoning. 

Pixel 8 Pro, the firm’s latest Smartphone model, will also run Gemini Nano to automate and improve various tasks, including features within messaging apps like WhatsApp. 

Upcoming plans will integrate Gemini models into Google Search, Ads, Chrome, and Duet AI. 

Google Gemini for Developers

Next week, on December 13th, Gemini Pro will become available to developers and enterprise clients through an API on the Google AI Studio or Google Cloud Vertex AI. 

They’re also launching AICore, a new developing platform for Android 14, which will include access to Gemini Nano for building on-device apps. 

However, Gemini Ultra is still in the safety check stage and will be made available for selected customers and partners before its full release, which is planned for next year. 

Gemini or GPT – Who’ll Dominate the Multimodal AI Space

While it’s too early to answer the question of who will come on top in a Google Gemini vs. OpenAI GPT4 facedown, it’s clear that the latter now has a fine contender for the throne after being pretty much the undisputed leader of the multimodal AI models. 

With the launching of Gemini, Google is making one of the most substantial steps into the AI product race yet and setting up to compete for the number one spot in the space. 

Who do you think will win? Share your thoughts!

THE AUTHOR

Ivanna Attie

All About Ivanna

I am an experienced author with expertise in digital communication, stock media, design, and creative tools. I have closely followed and reported on AI developments in this field since its early days. I have gained valuable industry insight through my work with leading digital media professionals since 2014.

• NETWORKING