Google announces the results of Gemini 1.0, surpassing GPT-4 in testing

About Google DeepMind who developed Gemini 1.0

Google has announced the merger of Google Brain and DeepMind to form a single team, called Google DeepMind. Let’s see the inside story of these teams.

Google Brain, founded in 2011, is Google’s deep learning platform that contributes to projects such as Google Translate and TensorFlow, an open-source software library for everyone to train their machine learning models.

DeepMind was created in 2010 and acquired by Google in 2014. DeepMind was working on several important projects, such as AlphaGo (Deep Learning AI).

The CEO of Alphabet and Google, Sundar Pichai, mentioned that “the combination of these two AI research teams will significantly accelerate efficiency and our progress in AI.”. He also referred to Demis Hassabis, CEO of Google DeepMind, saying, “He will lead the AI development team with new capabilities and come with AI security.”

What is Gemini?

Gemini is the largest and most capable AI model of Google DeepMind.

Google Gemini 1.0

Gemini supports various types of input data, including text, code, audio, images, and video. It is the most flexible model, which means it can efficiently run on everything from large data centers to mobile devices like smartphones. Google has published a video demonstration of the capabilities of Gemini. The video shows the interactions between human and Gemini by drawing pictures and asking Gemini about what it sees.

It shows that Gemini can correctly describe the drawing with detailed knowledge and express both text and voice naturally.

In addition to the drawing, when the tester picks up the object and compares it to the drawing, Gemini naturally expresses shock, and it can correctly predict what the object the tester picks up is. Moreover, it can also pronounce the object in other languages.

Just watching the beginning of the video, it is very interesting. However, Gemini’s capabilities do not end there. It can also create games for us with game answers and solutions. These are just some parts of the video. Anyone who watches until the end will be amazed by Gemini’s capabilities.

The first release of Gemini will be available in three different models:

3 models of Gemini
  1. Gemini Ultra: The largest and most capable model for highly complex tasks.
  2. Gemini Pro: The best medium-sized models for scaling across a wide range of tasks.
  3. Gemini Nano: The most efficient models are designed to run on devices like smartphones. The Google Pixel 8 Pro is the first smartphone that can use Gemini Nano.

Test results of Gemini Ultra outperforms GPT-4 in almost all aspects

When comparing the results of OpenAI’s GPT-4 with Google’s Gemini Ultra, Gemini beats GPT-4 in several areas. For example, Gemini Ultra scored 90% on the MMLU (Measuring Massive Multitask Language Understanding), or a set of knowledge and language understanding questions, while GPT-4 scored 86.4%.

In addition, Gemini also outperforms GPT-4 in mathematics and coding tests.

Gemini Ultra comparing with GPT-4

Gemini has a higher score than GPT-4 in almost every test topic in understanding image, video, and audio data.

Gemini comparing with GPT-4V

Currently, Google is providing Gemini 1.0 in English to people in more than 170 countries through Google products, and there are plans to support other languages in the near future.

Get ready, Gemini Ultra is coming soon

Google is testing the security and reliability of Gemini Ultra, and it will be available soon to selected customers, developers, partners, and security experts for early experimentation and feedback before rolling it out to general developers and enterprise customers.

Conclusion

AI and machine learning are continuously evolving, so we can see the capabilities of AI chatbots like Gemini 1.0, which will definitely help with human tasks. This is because Gemini can interact with humans naturally and it has knowledge that surpasses the knowledge and capabilities of humans. Gemini also has a large data source and can completely process information. In the future, we are likely to see even more intelligent AI, which could be enormously beneficial to humans in many ways.

Source:

https://deepmind.google/about/

https://deepmind.google/technologies/gemini/#introduction

https://blog.google/technology/ai/google-gemini-ai/?utm_source=gdm&utm_medium=referral#sundar-note

https://goo.gle/GeminiPaper

Image by Freepik