Google DeepMind has created a new Artificial Intelligence (AI) competitor to ChatGPT. This AI named Gemini can simultaneously understand different types of multimedia such as images, videos, audio and text and respond accordingly.
Most artificial intelligence technologies can only understand or generate one type of content. An example is chatgpt. OpenAI's AI can only understand information given as text. It can then respond accordingly, creating tailored content. If we talk about midjourney again, it can create pictures according to written instructions.
According to Google DeepMind's technical report, Gemini Ultra outperformed other AI models, including ChatGPT-4, in 30 out of 32 AI R&D benchmarks.
But Gemini is different. It can understand different types of content besides text. This is claimed in a recently published blog post by Google. The post was published on December 6.
Initially, Google released a total of three versions of Gemini 1.0. Among them, Gemini Ultra can handle the largest scale and most complex types of tasks. Gemini Pro is connected to Google's digital services. And Gemini Nano has been made suitable for use in smartphones.
According to Google DeepMind's technical report, Gemini Ultra beat other AI models, including ChatGPT-4, in 30 out of 32 benchmarks in artificial intelligence research and development. These include topics ranging from college level exams to ethics, science and technology, law.
According to Google DeepMind's technical report, Gemini Ultra beat other AI models, including ChatGPT-4, in 30 out of 32 benchmarks in artificial intelligence research and development. These include topics ranging from college level exams to ethics, science and technology, law.
In particular, Google's artificial intelligence models have successfully passed 9 image analysis, 6 video understanding, 5 audio and translation standards, and 10 text and logic understanding standards. However, the Gemini Ultra lost to the GPT-4 in the two tests of text comprehension and logic.
Building a model that can analyze multiple types of content is a daunting task. Because, in that case AI has to provide various types of data for training. Also the amount of data is huge. Efficiency therefore decreases. When it comes to correcting various types of errors, the AI is not able to improve much. At this time artificial intelligence models show 'overfit' characteristics. That is, it gives better results on the data it is trained with. But when given new type of data or instruction it can no longer complete those tasks.
Another thing is that in multimodal training, artificial intelligence is usually trained with different types of content at the same time. Then the model is made complete by combining everything. No such thing has been done in case of Gemini. Different types of content are provided together in the training dataset, i.e. the data provided for training. Google DeepMind scientists have used web documents, various books and codes to collect these data. But this training is given under human supervision. That is, the supervised learning model—in which a human tells the AI model where it is going wrong and how to correct it—is followed.
It took Google all the way up for this training. They used the famous Tensor Processing Unit or TPU for this very large scale work spread across their multiple data centers. Several thousand such TPUs—also known as AI accelerator chips—were used to train the Gemini model. As the name suggests, what this chip does is to speed up the work of artificial intelligence. Google also said so. Google DeepMind, its artificial intelligence research division, developed the chip primarily to speed up artificial intelligence training, Google said. Not only that, DeepMind built a cluster of 4096 chips called 'Superpod' to teach Gemini. As a result, Gemini is trained in much less time than before.
But it is still not a completely flawless artificial intelligence. Still it sometimes gives wrong information with 100% confidence. That is, this wrong information is considered correct
DeepMind's scientists have developed the Gemini AI model in such a way that it can be used even for immediate needs. For example, you are cooking food. At that time, he gave Gemini a picture to tell him what to do in the next step. Gemini can immediately follow this instruction.
But it is still not a completely flawless artificial intelligence. Still it sometimes gives wrong information with 100% confidence. That is, this wrong information is considered correct. This is called a 'hallucination'. The name is correct, to say it! However, this is Geminis biggest flaw. This is due to bias or various limitations in the data provided for training. Such errors are difficult to correct.
However, Gemini has become one of the pioneering AI models of today. It can help users in many ways as it is connected with Google services. Even though ChatGPT has been defeated by various criteria, it cannot be said that it has been surpassed at all. But you don't need to be a rocket scientist to understand that Gemini will bring new opportunities in the world of artificial intelligence in the future.