The world has been exploring the possibilities of artificial intelligence (AI) since the 1950s, a period of time in which the mathematician Alan Turing laid the foundations of this discipline with his article ‘Computing Machinery and Intelligence’.
Yet in seven decades of development, the widespread feeling that AI is advancing in leaps and bounds has never been stronger. One of those responsible for this reality has been the OpenAI conversational chatbot, ChatGPT.
The heart of this powerful tool is GPT-3.5, one of the autoregressive language models most advanced out there. And its capabilities are clear: it maintains conversations in natural language, understanding the context, and generates texts of all kinds.
If GPT-3.5 has surprised us, making us feel that feeling when we used the Internet for the first time, GPT-4, the next evolution of this language, promises to make a dramatic leap in its capabilities. And this evolution is on the way.
What is GPT-4?
GPT-4 is the presumed name of the next model of OpenAI pretrained language, an artificial intelligence research company that has previously come up with older versions of the model. GPT (2018), GPT-2 (2019), GPT-3 (2020) and GPT-3.5 (2022).
The capabilities of GPT-4 will be directly related to the language. You are expected to be able to perform tasks such as text generation, summaries, machine translation, answering complex questions, and much more with astonishing accuracy.
The most notable leap in the next version of GPT will come with its ability to respond like a human. Your answers and interactions will be more precise and consistent, so you will score points when facing the Turing test.
How will GPT-4 work?
GPT-4, like the previous models, by itself is a language with the potential to be exploited in different systems. GPT-3, for example, has been implemented in commercial applications such as Jasper Ai word processors and Canva Docs.
GPT-3.5, the most advanced version so far, has surprised us with ChatGPT. GPT-4, could go further from writing assistants, machine translation and chatbots, to voice assistants and even search engines (Microsoft Bing is a candidate).
Internally, GPT-4 will be trained with datasets with large amounts of data that will serve to learn and generate language similar to that used by humans. Behind the model is a processing technique known as “Transformer”.
The objective of this architecture, presented by Google in 2017, is to innovate in the implementation of layers that allow adapt model to be effective and efficient in different tasks. OpenAI, in its GPT models, has used it to implement various layers.
The Transformer architecture, through its layers, converts each word into a numerical value that allows the model to process the text mathematically, is responsible for processing it through a neural network and “pays attention” to understand it.
But it’s not just about layers. In the GPT models there is also a lots of parameters. These are shaped during the machine learning process and are theoretically directly related to model performance and accuracy.
How is it different from GPT-3?
According to the OpenAI documentation, GPT-3 has 12 layers and 175 billion parameters. The main difference between the latest OpenAI model and its evolution, according to Wired, will be in the parameters. GPT-4 will have 100 trillion parameters, almost 600 times more than its predecessors.
It should be noted that OpenAI has remained silent thus far. However, they also point out that the evolution of the model could not necessarily be achieved increasing the number of parameters. A major limitation would be that the computing power needed to grow in this regard costs millions of dollars per hour.
When will GPT-4 be released?
OpenAI quietly launched GPT-3.5 in November 2022 with the release of ChatGPT. There’s no official roadmap for the arrival of GPT-4, or whatever OpenAI decides to call the evolution of the model, but TechCrunch notes that it could arrive sometime in 2023.
Images: Aideal Hwa |rawpixel.com | Open AI | Screenshot ChatGPT
In Xataka: I’ve played a role-playing game with ChatGPT and a movie worthy of an Oscar has been made
George is Digismak’s reported cum editor with 13 years of experience in Journalism