Open AI may introduce GPT-4 neural network in mid-March
At the end of 2022, OpenAI launched the ChatGPT bot, which can write text on any request. For example, you can ask him to write an essay, compose a poem or a story, come up with code in one of the popular programming languages like Python, or simply ask a question – he will answer it without any problems. The ChatGPT bot still surprises us with its capabilities and the head of Microsoft, Sundar Pichai, even called it “a more important invention for mankind than fire and electricity.” But everything that ChatGPT does today is clearly not the limit of its capabilities. Already in mid-March, the GPT-4 neural network can be presented, thanks to which the bot will clearly become faster and smarter. There are already a lot of rumors around the upcoming update: they say that a new neural network can be trained on 100 trillion data, be able to create images, and so on. Is it true?
When will the new version of the GPT-4 neural network be released?
The fact that Open AI is developing an improved GPT-4 neural network is already known to everyone. Representatives of the organization do not deny this, but on the contrary, share some details. For example, the head of Open AI, Sam Altman, reacted to a rumor about training a new neural network based on 100 trillion data by calling it “complete nonsense.”
Microsoft Germany CTO Andreas Braun recently announced that GPT-4 could see the light of day as early as mid-March. According to him, the new version will include “multi-modal models that will offer completely different possibilities.”
Perhaps the GPT-4 neural network will be presented on March 16
There is nothing surprising in the fact that such loud statements are made by representatives of Microsoft. The fact is that the creators of the Windows operating system are the main investors of Open AI – they invested a fabulous $ 10 billion in the development of the GPT neural network. Rumor has it that the latest GPT-4 is already built into the updated Microsoft Bing search engine, but not everyone can use it – access to millions of users was promised to open before the end of February, but this has not happened yet.
Updated Bing search engine with built-in neural network
So the GPT-4 neural network can be presented on March 16, 2023. But whether it will be immediately available to a wide range of people is not clear at the moment. There is a possibility that the neural network update will follow in the footsteps of GPT-3 and at first be open only to Open AI partners and academic institutions. But there is also a good chance that it is indeed already built into the Microsoft Bing search engine and will form the basis of the ChatGPT bot.
Хотите пользоваться нейросетью ChatGPT, но не знаете как? Читайте нашу статью «Нейросеть ChatGPT пишет тексты на любую тему — как ей пользоваться».
Capabilities of the new neural network GPT-4
Most of all, people are not even interested in the release date of GPT-4, but in its abilities. Immediately it is worth warning that before its release, you need to moderate your expectations. The authors of Forbes believe that there will be no excitement like in the case of the launch of the ChatGPT bot – at first people were amazed that the computer learned to write essays and poems, but now few people will be impressed that he learned to do it a little better.
The neural network will be better at writing text and answering questions
Judging by what they say on the Internet, the neural network really will not be trained on 100 trillion data. Some sources say that if GPT-4 gets access to more information, then we will only talk about 17 trillion parameters. But this will also be a huge step forward, because the current version of the GPT neural network is trained on 175 billion data.
After the release of GPT-4, the ChatGPT bot can start working faster and without errors
But this does not mean that the new version of the neural network for writing texts will not receive noticeable improvements. It has been empirically proven that the most important thing in this kind of neural networks is not the amount of information in the “brain”, but the ability to manage it. For example, the competing language model Megatron 3 is trained on a much larger amount of data, but cannot give the same smart answers as the OpenAI neural network. If developers can clean up the structure of GPT-4, it will start to work faster and give more accurate answers – users will clearly appreciate this.
ВАЖНО: считается, что в будущем нейросеть для написания текстов ChatGPT будет запрещена в школах. Но некоторые специалисты считают, что сделать этого не получится. Если интересны подробности, читайте нашу статью «Почему нейросеть для написания сочинений ChatGPT не получится запретить в школах».
The neural network will write code better
There is reason to believe that the GPT-4 neural network will write program codes even better than the current version. The fact is that relatively recently, OpenAI hired programmers who can explain in human language what the code they write does. Perhaps thanks to the hard work of new employees, GPT-4 will be able to write even better code for requests like “write a calculator in the Python programming language.”
The ChatGPT neural network even now does a good job with programming, but there is no limit to perfection
GPT-4 neural network will learn to draw?
There were rumors that Open AI would combine the capabilities of the language model and the Dall-E 2 neural network in GPT-4 to draw images on a text request. If this were true, the ChatGPT bot could not only generate text, but also supplement it with diagrams, graphs, and other graphic elements. Alas, the head of Open AI, Sam Altman, assured everyone that GPT-4 will remain an exclusively text model.
GPT-4 neural network is unlikely to learn how to draw
Вам будет интересно: 5 нейросетей для создания уникальных фотографий и рисунков
Ultimately, here is what is known from the GPT-4 neural network:
- the update may be released in mid-March 2023, but it is not known in what form and to whom it will be available;
- it may already be built into the updated Bing search engine from Microsoft, for which you need to queue up to test;
- developers will not greatly increase the amount of data available to him, but will teach him to work better with what is;
- the GPT-4 neural network can learn to write code better;
- the GPT neural network will remain a language model and will not supplement texts with pictures.
It is worth emphasizing that almost everything written above is based on rumors from various sources. We will be able to find out exactly what the GPT-4 neural network is only after its release. Most intriguingly, in his keynote, Microsoft spokesman Andreas Braun noted that the new neural network will support video. This statement contradicts Sam Altman’s claim that GPT will remain a text-only model.
Как вы думаете, каким будет нейросеть GPT-4? Присоединяйтесь к обсуждению в нашем Telegram-чате.
Be that as it may, if something interesting happens, we will definitely tell you about it. So subscribe to our Zen and Telegram channels so you don’t miss anything important.
