Google launched Gemini on Wednesday (6), an artificial intelligence (AI) model treated by the company as the most powerful ever created by its team. It can already be tested on Bard (ChatGPT’s competitor), but for now, only in English.
According to Google, Gemini will be useful for tasks that require more reasoning ability. It can be used to help programmers create complex codes and students to do homework, for example.
In a demonstration, Gemini was able to analyze a photo of a physics exercise and identify an error in the calculation. The AI could present the correct solution and create similar problems for the user to solve.
Gemini surpassed human capacity in a knowledge and problem-solving test that combines 57 subjects, including mathematics, physics, and history, according to Google. The company also says that it performed better in this test than GPT-4, OpenAI’s model used in ChatGPT.
Learn more about Google’s new artificial intelligence:
- How does Gemini work?
- When will it be available?
- What makes Gemini more powerful?
- Is Gemini reliable?
How does Gemini work?
Gemini is the most flexible AI model ever developed by Google. The idea is that it can run on both large infrastructures, such as data centers, and more limited devices, such as mobile phones.
To achieve this goal, the new artificial intelligence will have three versions:
- ⚡ Gemini Ultra: larger and more powerful, focused on highly complex tasks – will be released only in 2024;
- 💻 Gemini Pro: designed to run a wide range of tasks and serve users, starting from this Wednesday on Bard, and developers, starting from December 13;
- 📱 Gemini Nano: created for mobile devices, it will run directly on the device, allowing it to function even when there is no internet – available on the Google Pixel 8 Pro, Google’s phone, to create audio summaries and suggest intelligent responses on WhatsApp.
When will it be available?
The first contact for many users with Gemini is expected to be with the Pro version, which will improve reasoning, planning, and understanding on Bard. From this Wednesday, Google’s robot will already use the new AI in English responses (other languages will receive the update soon).
In early 2024, Google will launch Bard Advanced, which will make Gemini Ultra available on Bard. And, according to the company, products like Search and Chrome will integrate with Gemini in the coming months.
From December 13, developers and companies will be able to use Gemini Pro through the Google AI Studio and Google Cloud Vertex AI artificial intelligence platforms. And Gemini Ultra will undergo further security checks before being released to everyone in 2024.
What makes Gemini more powerful?
Google says that the approach to developing Gemini was a bit different. From that adopted in other multimodal artificial intelligence models. Those that can understand different formats simultaneously.
According to the company, creating AI models usually involves separate “training” with a focus on each format. In a second stage, text learning is combined with image learning, for example, and everything is “packaged” into a single product.
According to Google, this method allows for some tasks. Such as describing images, but has limitations in cases that require more reasoning.
“We designed Gemini to be natively multimodal, pre-trained from the start on different modalities. Then we refined it further with additional multimodal data to enhance its effectiveness,” said Demis Hassabis, CEO of DeepMind, Google’s AI subsidiary.
Is Gemini reliable?
Google admits that, like all artificial intelligence, Gemini is also subject to the hallucination problem. An error that makes a robot’s response include incorrect, biased, or nonsensical information.
Gemini is capable of making the best safety assessments of an AI ever created by its team. The goal is for it to be able to avoid violent or stereotyped content, for example.
The creation of Gemini also involved training. What should not be shown to users from a set of 100,000 sentences considered toxic taken from the internet. A group of independent experts also works to test the limits of the new AI.
“We are approaching this work boldly and responsibly,” said Sundar Pichai, CEO of Alphabet, Google’s parent company. “This means being ambitious in our research and seeking capabilities that will bring enormous benefits to people and society, while building safeguards and working collaboratively with governments and experts to address risks as AI becomes more capable,” he said.
1 comment