Google’s Next Big Thing In AI-Model Gemini Outperforms ChatGPT


Google has officially announced its latest AI model Gemini that can now behave like a human, which may spark a debate about how technology could potentially benefit and affect human tasks. The race to render the human brain redundant has seen a significant advancement with Google’s latest multimodal model, which can handle a mix of inputs and has outperformed human experts in critical benchmarks.

This advancement surpasses other foundational AI models in specific tasks, including OpenAI’s GPT 3 and Meta’s LlaMA 2. The next-generation AI model powers Google’s various AI features and services, including Google Assistant. It is more advanced and can deal with sophisticated scenarios compared to previous-generation AI models.

This follows the latest generative AI systems released by companies such as OpenAI, Meta, Amazon, Anthroics, and Mistral. Reportedly, more than 50 organizations are developing artificial generative intelligence. Google has been an “AI First Company” for nearly a decade. Let us take a closer look at Gemini’s underlying technology.

What is Google’s Gemini?

This is the company’s largest and most capable AI model, taking another leap forward in the rapidly growing industry of artificial general intelligence with the announcement of Gemini. It is a more reliable and scalable model to train, as it is most efficient for its generative AI, which rapidly evolves the relative strengths of competing models over time.

Gemini is an advanced generative AI model trained to mimic human beings and outperforms GPT-4. It will power Google’s AI services and features going forward. Compared to other AI models that are built and trained for specific tasks, such as OpenAI’s DALL-E for image generation, GPT for text generation, and Whisper for audio, Google’s Gemini is built from scratch to process text, photos, and videos simultaneously, making it multimodal.

Capability Benchmark Gemini Ultra GPT-4
General MMLU (Representation of various questions in 57 subjects) 90.0% 86.4%
Reasoning Big-Bench Hard (Challenging tasks requiring multi-step reasoning) DROP (Reading comprehension) 83.6%   82.4% 83.1%   80.9%
Math GSM8K (Basic arithmetic manipulation) MATH (Challenging math problems) 94.4%   53.2% 92.0%   52.9%
Code HumanEval (Python code generation) Natural2Code (Python code generation) 74.4%   74.9% 67.0%   73.9%

It has been highly optimized for scientific breakthroughs and is adept in math and physics, benefiting researchers, students, and others. However, ChatGPT Plus is comparable to the GPT 3.5 model but is less capable than Gemini Pro, which provides access to the GPT 4 AI model.

Under the hood, the company’s AI research and implementation of safeguards are done collaboratively with the government and experts to avoid risk factors. This benefits users while presenting a challenge to Meta AI and OpenAI, which are among the leaders. Gemini was developed by Google’s DeepMind’s Brain Division and trained on the open web.

It includes the chatbot that has boosted the AI industry since the release of ChatGPT. Building safeguards and working collaboratively with the government and experts to address risks as AI becomes more capable is not necessarily a giant leap for the field as a whole. Furthermore, Gemini runs on an AI-optimised infrastructure using Google’s in-house-designed Tensor Processing Units (TPUs), which are more efficient than GPUs.

However, it is subject to shortages of TPUs, unlike GPT-4 and other models. Additionally, the TPU v5p Soc offers better performance for the price than the TPU v4 (2021), although the company did not provide performance information compared to Nvidia.

Capabilities of Google’s Gemini AI Model

Let us talk about some basic capabilities of the Gemini AI Model. The company describes its AI model as the most capable, flexible, and general model yet conceived.

  • Users can interact with hand gestures, images, drawings, etc., using this multimodal approach to take on complicated tasks like dialogue, solving visual puzzles, logical and spatial reasoning, and more. It can understand and operate across different combinations of input data.
  • It can refine datasets from multiple sources, extract relevant information based on user input, and create diagrams.
  • Gemini also has math and physics skills, which can help students, researchers, and others improve and understand their homework.
  • It has been updated with AlphaGO 2, which brings advanced coding capabilities such as writing, debugging, and understanding code in different languages, including popular ones like Python, Java, C++, and G. This will be helpful for prototyping web apps, quick coding, and solving competitive programming problems. Reportedly, Gemini can solve competitive coding problems better than 99.8% of participants in coding contests. Its coding proficiency was tested on HumanEval, where it solved 74.4% of the tasks. This also makes it suitable for coding, complex math, and theoretical computer science.

Since the launch of OpenAI’s ChatGPT a year ago, Google has declared a red alert and started developing an advanced AI model. They have also merged the company’s AI divisions to compete against Microsoft-backed OpenAI’s ChatGPT, DALL-E, and Whisper.

Google Gemini AI models

Gemini 1.0 was officially announced with support in English. However, Gemini AI will soon be supported in many other languages. This is said to be the first realization of the vision when the company formed Google’s DeepMind by merging DeepMind and Brainteam earlier this year. Google’s DeepMind has also made significant contributions to the development of Gemini. This AI model comes in three different sizes, optimized for different tasks.

Here are the details of Google’s Large Language Model (LLM). Although Gemini is still in its experimental phase, efforts have been made to improve the AI model and expand its capabilities and availability. Google runs everything from its Google data centers, and there are three different versions of Gemini optimized for different tasks: Ultra, Pro, and Nono.

Gemini Nano

It is a sophisticated version of Gemini, known as Nano, and it is said to be the most efficient. It can process data on-device and works even offline through on-device tasks. Via AiCore, which is a new system capability available on Android 14, the company has started rolling it out on Pixel 8 Pro devices.

Among the Gemini Series, it is the lightest AI model designed to run on-device locally. Android developers will soon be able to build the Gemini Nano. This advanced model can be run on mobile devices, vastly reducing the computing cost. Since these AI models can run directly on the mobile device for the first time, they will run natively on the device rather than in the cloud on a server, which ensures that private data is restricted on the device.

Gemini Pro

To incorporate the Gemini AI model into Google’s AI chatbot called Bard, it is designed to handle most basic tasks. From today, Gemini Pro is powering some of the Google products, including the Google Generative Search Experience, Ads, Chrome, and Duet AI. Through Google’s data centers, It is designed to deliver faster response times and understand complex queries.

Gemini Ultra

One of the most advanced AI models, scheduled to be released early next year, is among the largest and most capable models for dealing with complex tasks. Google plans to release the Ultra through Bard early in 2024. It excels in image benchmarks, native multi-modality, and complex reasoning abilities.

This AI model can simultaneously understand text, photos, and videos, providing a multimodal chatbot experience. Furthermore, before Google makes it official, Gemini Ultra is currently undergoing fine-tuning and reinforcement learning from human feedback (RLHF).

It is offered to selected customers, developers, partners, and safety and responsibility experts for preliminary testing and feedback. On the other hand, Gemini Ultra has exceeded the Large Language Model (LLM) research and development on 30/32 widely used academic benchmarks in text and multimodal output. It scores 90% in MMLU (Massive Multitask and Language Understanding), which includes 57 subjects like math, physics, history, law, medicine, and ethics.

Gemini Ultra Demo

Check out the following six-minute video demo of Google’s most advanced LLM model, Gemini, which unveils the capabilities to generate game ideas, connections between objects, and much more. It can keep a child engaged for hours. It can also help users understand different languages, including Spanish, French, Korean, etc.

This gives us a glimpse of Gemini’s multimodal capabilities of generating text, analyzing images, and much more.The demo showcases scanning a handwritten worksheet of mathematical formulas, marking errors, and explaining them. This AI model is going to benefit everyone.

During the demo, the company also highlights Bard, which helps a mom and her son with homework to determine mistakes. This shows how well the model can analyze images and follow the inherently multilingual training in more than 100 human languages.

How to Use Google’s Gemini

Google started rolling out the Gemini AI model in phases through its Bard and Android 14 feature drops to the Google Pixel 8 Pro.

So, if you want to try Google’s latest AI model, Gemini, here are the following instructions on how you can use it: However, the Bard upgrade will not be released in the European Economic Area, which includes the EU, Switzerland, the UK, and Europe, as Google seeks clearance from regulators.

Google Bard with Gemini

While the Bard was relaunched earlier this year, it is now upgraded with Gemini to take a shot against OpenAI’s popular ChatGPT. Powered by Gemini, the latest large language model is available in more than 170 countries and territories, including India.

It is now capable of reasoning, planning, and understanding for free, whereas competitors like OpenAI have made GPT-4 available via subscriptions, which cost $20/month.

  1. Open the Google Bard web app, for which visit
  2. After that, you need to log in with your personal Google account.
  3. In the text field, start writing the prompt, and then hit enter to get the output.

Bard Advance with Gemini Ultra

As Gemini Ultra is set to be released early next year, Google will also upgrade to Bard Advance. Initially, Bard Advance will be available to selected customers. Reportedly, Google’s Bard Advanced will be capable of handling many tasks like processing images, text, and videos simultaneously.

Android 14 Update

Google pushed a December Pixel Feature Drop to the Google Pixel 8 Pro that is intuitive and can handle tasks better.

Smart Reply

It suggests a significant upgrade in the messaging application, which helps users generate content using natural language processing, assisting users in getting relevant and natural responses.

  • It brings automatic replies to messaging services like WhatsApp through Gboard (formerly Google Keyboard).

To start using it, you need to follow the following instructions: You need to enable AiCore (Enable AiCore Persistent) from the device’s developer options to start using Smart Reply on WhatsApp.

Once you have enabled AiCore, Gemini Nano helps users with Smart Reply Suggestions in GBoard within GBoard.

Summaries of Call Recording and Voice Recording in the Recorder App

Google’s Recorder App already has one of the most advanced transcription functionalities, which helps users transcribe voice recordings, including call recordings or voice recordings on the Recorder app. Gemini Pro on the device can now help users generate summaries that include a quick overview of main points and highlights.

  • Users can now get summaries of call recordings and voice recordings.

Some notable features that you can start using on your Google Pixel 8.

Google plans to upgrade its products and services with Gemini.

Over time, it will eventually be infused into Search, YouTube, Workspace, Chrome, Duet AI, and other products and services.

However, the Search Generative Experience, which tries to answer search queries with conversational-style text (SGE), is not widely available. Compared to other advanced AI models to date, Gemini Ultra stands out with the release of Gemini Ultra and its native multimodal characteristics. In contrast, other models like GPT-4 rely on plugins and integration for such multimodal functionality.

Google will monetize Gemini by offering the AI model to developers and businesses.

Developers and businesses will be able to leverage the Gemini AI model by integrating it into their products and services through the Gemini API starting December 13. The API offers incredible rapid prototyping and app development benefits when handling multimedia content.

To run Gemini Ultra, which can handle complex tasks and requires significant resources, will cost more. This means that Gemini Ultra will likely underpin the paid AI offering from Google. Google plans to license Gemini to customers through Google Cloud for use in their applications while maintaining the power of consumer-facing Google AI apps like Bard Chatbot and Search Generative Experience.

Google will allow developers, businesses, and enterprise customers to access Gemini Pro for their products and services via the Gemini API. Development can start in Google’s AI Studio and Google Cloud Vertex AI starting December 13. Additionally, Android developers will have access to Gemini Nano via AiCore through an early preview forum.

Google to take on OpenAI’s GPT-4

AI competition has rapidly accelerated in the past few years, with OpenAI’s most advanced AI model, GPT-4, offering promising generative AI. Google claims that Gemini has 5x the computational power of GPT-4, which makes Gemini faster. After the launch of OpenAI’s ChatGPT, Google declared a code-red and has since gathered all its resources for AI development.

Concerns about Gemini AI

Many experts and users have shared their concerns about the Gemini AI model, as it could cause loss, destructive behavior, misinformation, and even trigger the development of nuclear weapons. However, there are still some limitations, including hallucinations. The company has implemented several safeguards to address all concerns while developing the Gemini AI.

Also, it works closely with the government and experts to address the risks of AI as it becomes more capable. Not to mention the potential enormous benefits of GenAI to people and society. The shares of Google Parent Alphabet were down 0.7% in intraday trading on Wednesday but have gained approximately 46% so far this year.

Leave a Comment