What You Must Know About Google's Gemini

Google CEO Sundar Pichai says Gemini will eventually be integrated into Google's search engine, its ad products, the Chrome browser.

Google made headlines on December 6 as it launched Gemini, touted as its most powerful artificial intelligence (AI) model, trained to behave in human-like ways.

Competing with OpenAI's ChatGPT, the race for AI supremacy is clearly set to get fiercer.

Google CEO Sundar Pichai first teased Gemini at the I/O developer conference in May and called it a huge leap forward in an AI model that will ultimately touch and impact all of Google's products.

Here's what you need to know:

Is Gemini a single model, or does it have versions?

Gemini comes in three versions: Pro, Ultra, and Nano.

Gemini Nano, the lighter version, is meant to run natively and offline on Android devices.

Gemini Pro, the heavier version, will soon power several Google AI services and is the backbone of Google's chat-based AI tool, Bard.

Gemini Ultra is the most powerful large language model (LLM) Google has yet created and seems to be mostly designed for data centres and enterprise applications.

Which languages does it support?

Gemini is currently available only in English, but support for other languages is expected to come soon.

Pichai said the model will eventually be integrated into Google's search engine, its ad products, the Chrome browser, and others globally.

How does Gemini compare with GPT-4?

Google conducted several benchmark tests comparing Gemini with GPT-4. Google claims that Gemini achieved a remarkable score of 90 per cent on the massive multitask language understanding (MMLU) test, surpassing human experts (89.8 per cent) and outperforming GPT-4 (86.4 per cent).

MMLU uses a combination of 57 subjects, such as mathematics, physics, history, law, medicine, and ethics, to test both world knowledge and problem-solving abilities.

However, it is important to note that Google used different prompting techniques for the two models. GPT-4's score of 86.4 per cent relied on the industry-standard '5-shot' prompting technique.

In contrast, Gemini Ultra's 90 per cent result was based on a different method -- 'chain-of-thought with 32 samples'.

A lease of life for existing Google products

Google's chat-based AI tool Bard will be a key beneficiary of Gemini, as it is using a tweaked version of Gemini Pro for more advanced reasoning.

A second version of the AI chatbot called Bard Advanced is likely to follow early next year, giving access to the company's most cutting-edge models, including Gemini Ultra.

Experts believe this too will be locked behind a subscription, just like ChatGPT Plus.

Google's Pixel 8 Pro is also expected to benefit from the Gemini Nano, bringing more on-device AI features.

What lies ahead?

As Gemini integrates into more Google products and services, its power and potential to solve real-world applications and problems will gradually become clearer.

We have to watch out for deepfakes or misleading information, along with the serious issue of LLM hallucinations, which are an increasing problem with its competitors.

Moreover, both Google and its users will have to keep up with the rapidly changing tech regulations, adapt quickly to the ever-evolving world of AI, and really think about the ethical side of using such a powerful and versatile AI tool.

Whether it's a giant leap from Google or a leap of faith, time will tell, but for now, it's surely the show-stopper!

Feature Presentation: Ashish Narsale/Rediff.com