What is GPT-3

OpenAI, the Californian company co-founded by Elon Musk dedicated to artificial intelligence research, has created (with an approximate investment of $4.6 million) a new AI language that promises to be a revolution.

Just a few months ago, OpenAI presented the new artificial intelligence language GPT-3 , which is a Deep Learning model, a Machine Learning subtechnique made up of algorithms that are trained to recognize patterns in data and learn through them. examples, that is, that GPT-3 is composed of a recurrent neural network with long-term memory .

GPT-3 has arrived stomping and leaving the technological community speechless, and it seems to be one of the most powerful artificial intelligence models to date. And it's only in beta version.

As an artificial intelligence language , GPT-3 is capable of generating texts through a single statement and “filling in” the rest of the writing by processing more than 175,000 million parameters. A very relevant fact considering that its previous version, GPT-2, processed 1.5 billion parameters and was presented in 2019. That is, in just one year the evolution has been enormous.

GPT-3 was trained with data from: Common Crawl, an organization that has been collecting all public information on the Internet for years; WebText2, a collection of text extracted from different web pages that has been polished by people; from books; and from Wikipedia.

It is also capable of translating texts into other languages in a matter of seconds, although in English it works much better, for the moment.

Furthermore, one of the new functionalities of this AI is that it can adapt the text to different writing styles (journalistic, novelistic, formal, etc.) to give the text a less “artificial” character. Although it should be said that at the moment GPT-3, although it is a semantic intelligence that can generate very coherent texts, is not capable of differentiating whether the created text makes sense or not, it does not understand what it writes. Will GPT-4 be able to do it?

GPT-3 also programs

Another thing that has amazed many programmers is that, although it may seem like science fiction , GPT-3 is capable of programming . And for example, a button:

This is mind blowing.

With GPT-3, I built a layout generator where you just describe any layout you want, and it generates the JSX code for you.

WHAT pic.twitter.com/w8JkrZO4lk

— Sharif Shameem (@sharifshameem) July 13, 2020

In this tweet published by Sharif Shameem, CEO of Debuild, you can see how with GPT-3 you can program a web page without needing to know code. Obviously, GPT-3 cannot replace a computer programmer, but it can help you speed up the process of creating your projects and save time on some tasks. If you are thinking that an AI like GPT-3 can leave you without a job... don't worry, at the moment artificial intelligence has not yet matched human intelligence and creativity!

We are convinced that new technologies like GPT-3 will not destroy workplaces, they will transform them into others that today we cannot even imagine. Digital transformation aims to automate processes to save time and unnecessary procedures for humans, and we believe that artificial intelligence will achieve this.

The origin of GPT-3

Before the revolutionary GPT-3, OpenAI, the company of Elon Musk and Sam Altman founded in 2015, worked on other versions of GPT, the first of them in 2018, and other artificial intelligence projects such as OpenAI Gym, a tool for research of reinforcement learning, and Universe, an AI training software.

But the true origins of GPT-3's artificial intelligence date back many decades. You could say that its origins lie in the 60s , when Frank Rosenblatt created the Perceptron, the first artificial neuron. Perceptron marked a radical change in machine learning. Although it was not until the 1980s when Rosenblatt's research was recovered following the rise of research in neural networks to transform them into layers with multiple interconnected artificial neurons.

At the same time, at the end of the 60s, the first Chatbot also appeared, ELIZA, who was capable of “maintaining a conversation” with a human in a coherent manner. And although it was only a code, it developed what we know today as the “ELIZA effect”, an effect of captivation towards technology.

Thanks to all this research at the end of the 20th century, projects have been developed that we already use daily and without which it would be strange to live, such as Siri or Cortana recommendations, the recognition of Facebook friends, the organization of photos by “faces”. ” of smartphones or the new Tesla cars.

But are we under the “ELIZA effect” with GPT-3?

If GPT-3 fell into the wrong hands...

This technology is intended to become a tool to help people, but it could also be used for not very ethical purposes. Well, as many artificial intelligence specialists have warned, great care must be taken with possible “fake news” that could be generated in seconds and flood the network to carry out a disinformation campaign without many people realizing what happened. . And thousands of ads could even be created for malicious purposes.

And even if you think that you could recognize a text written by GPT-3, the results of a study carried out by Adrian Yijie Xu are surprising, which reveal that only 52% of readers can assure that the texts produced by GPT-3 are artificial. Therefore, there is a large percentage of the public that can read fake news and treat it as true, thus forming a loop of massive misinformation.

With GPT-3 available to anyone, the credibility of the content on web pages would be seriously jeopardized.

Without a doubt, GPT-3 is a revolutionary artificial intelligence , but not so much because of its capabilities, but because of its potential. Since seeing where we have managed to get, you can guess where we could go in a few decades.

At the moment this technology is the property of OpenAI and no other company has access to it. Therefore, it is possible that new similar technologies from other competing companies will appear very soon.

And now, do you think this text was written with GPT-3?