This week, Google introduced "Gemini," a formidable competitor to OpenAI's GPT-4.

Gemini comprises three distinct models varying in size and capabilities. As per Google's announcement, its most advanced version, Gemini Ultra, is geared for intricate tasks and surpasses GPT-4 in various domains, such as historical knowledge, legal expertise, Python code generation, and multi-step reasoning.

Google highlighted Gemini's outperformance of GPT-4 on the Massive Multitask Language Understanding (MMLU) test, a widely recognized benchmark assessing AI models' knowledge and problem-solving abilities.

Gemini Ultra, although not yet accessible to the public, promises significant advancements in AI capabilities.

Gemini, A Formidable Rival of GPT-4

Kevin Roose said in a podcast on Hard Fork that he likened the Massive Multitask Language Understanding (MMLU) to the "SATs for AI models."

The MMLU encompasses 57 subjects, spanning math, physics, history, law, medicine, and ethics, assessing broad knowledge and problem-solving skills.

Gemini Ultra achieved a 90% score on the MMLU, surpassing GPT -4's 86.4%, marking an essential milestone as the first model to outperform human experts, who scored approximately 89.8%.

This achievement led Roose to speculate on artificial general intelligence (AGI).

While GPT-4 excelled in everyday common sense reasoning, Gemini Ultra was praised for its natively multimodal design, enabling seamless processing of various data types-text, audio, code, images, and video-without the need for combination from separate models.

Google claims Gemini's built-in multimodality offers superior input comprehension compared to other multimodal models.

Additionally, Google suggests that Gemini's architecture and computing power could outperform GPT-4, a point echoed by analysts at SemiAnalysis.

The Generative AI Market Battle

The generative AI market is projected to explode, reaching $1.3 trillion by 2032 from $40 billion in 2022.

One source exclaimed their forecasts that generative AI software will generate $280 billion in revenue by 2032. It is pivotal to Google's Gemini models, fueled by its own Tensor Processing Units.

Gemini Pro launches on Google Cloud, followed by Gemini Ultra after safety checks. These models boost Google's cloud business and refine applications like Bard.

Integrating Gemini into products like Search, Ads, and Chrome is strategic in facing AI's impact on search. Gemini Nano aids Android 14 smartphones, enhancing on-device AI features in Pixel 8 Pro devices.

Using custom AI chips over Nvidia GPUs, Google gains a cost advantage amid the AI surge, which is crucial in training advanced generative AI models.

What Exactly is Google's 'Gemini?'

In AI, Google's Gemini shows up as a model that can handle text, images, and sound simultaneously, giving it an advantage over GPT-4, which works with words. It says it can solve problems like a human, which makes people think AI can deal with complicated stuff no one thought it could do before.

But is Gemini really going to do what it claims? Is it better than GPT-4? There are still concerns about powerful AI and what could go wrong.

The start of Gemini means a big moment for AI stuff. Folks are curious about what kind of effect it might have. The writer invites people's thoughts on how Gemini could change things and wants discussions on where AI is headed.

As the journey for AI keeps going, the questions stay open without answers for now until later on down the road. The article ends by asking readers to share and support it, realizing that getting people involved helps spread the word further.