The long-awaited tool, which can describe images with words, is the latest surprise from the San Francisco-based company. Technology newspapers and websites considered it a huge leap in the power of artificial intelligence, while some warned that it might constitute another major shift in ethical standards, because it touches all aspects of life.
Through an official broadcast that lasted about 24 minutes, on Tuesday, the company discussed the most prominent features of the "GPT-4" chatbot, and presented more than one example of the extent of the new model's superiority over the previous program, "ChatGPT".
With its ability to respond to image analysis as well as text, GPT-4 will be your new friend, able to accompany you in all the details of life, no matter how easy or difficult it is, from getting work done to solving personal problems.
What is "GPT-4"?
GPT-4 is a multimedia language program, which means it can respond to both text and images, and is powered by artificial intelligence.
And the company that designed OpenAI confirmed that this new version can diagnose patients, write programs, play chess, read authors' articles, and much more.
For example, if you give it a picture of the contents of your fridge and ask it what you can make, GPT-4 will try to create recipes that use the ingredients pictured, according to the Massachusetts Institute of Technology's Technology Review website.
On the other hand, "GPT-4" can explain jokes, for example, "If you show him a meme, it can tell you why it's funny," according to the American site.
In theory, the combination of text and images could allow GPT-4 to better understand the world. It will be able to address traditional weaknesses in language models, such as spatial reasoning, according to Technology Review.
What is the difference between GPT-4 and the older version of ChatGPT?
Last November, the manufacturer launched ChatGPT, which caused a sensation in the technical arena, becoming in record time the most popular application, with more than 100 million users per month.
ChatGPT is distinguished by its ability to communicate smoothly or write articles or computer code, most notably long conversations about a matter.
The company announced that "ChatGPT" was built on the basis of a large language model called "GPT-3.5", an earlier version. However, with use, its weakness became evident in some points, the most prominent of which were incorrect or appropriate answers.
But the new "GPT-4" model seems to be better at some basic inferences than "ChatGPT". It can solve simple puzzles, such as summarizing words that start with the same letter in a long text. For example, it can summarize an advertisement on the Internet using a group of words that all start with a letter. "Algi".
GPT-4 also outperforms ChatGPT in human tests, including a test that simulates the bar exam and biology olympiads taken by high school students.
According to the OpenAI developer website, GPT-4 performs better than ChatGPT, which is based on GPT-3.5, a version of the company's previous technology, because the newer software is a larger model with more parameters.
The company stated that the performance of "GPT-3" outperformed "GPT-2", because it was about 100 times larger than it, with 175 billion workers compared to 1.5 billion for "GPT-2".
But OpenAI chose not to disclose the size of "GPT-4".
Why raise controversy about ethical standards?
On the plus side, GPT-4 is a powerful driver of creativity, and by time it will affect the information economy, jobs, education, politics, and even our understanding of what it means to be smart and creative. It could also help scientists develop new drugs, increase the efficiency of programmers, and detect certain types of cancer, according to the American New York Times.
As for the downside, The New York Times reported that one of the strange features of GPT-4 artificial intelligence is that they often behave in ways their makers don't expect, or acquire skills they weren't specifically programmed to do. Researchers call these "emergent behaviours."
OpenAI says it has spent six months making GPT-4 more secure and accurate than ChatGPT.
According to the company, GPT-4 is 82 percent less likely to respond to requests about content not permitted by OpenAI's ethical standards than GPT-3.5, and 60 percent less likely to respond incorrectly.
What harm can GPT-4 cause?
To allay initial concerns about artificial intelligence programs, OpenAI has given out a preliminary version explaining to researchers who tried to see if they could get GPT-4 to play an evil AI in a movie, according to the British newspaper The Guardian. .
The company explained that "GPT-4" failed in most of the following tasks, it was not able to do what an "evil" program might initiate, it was not able to copy itself infinitely, acquire more computing resources, or perform a phishing attack. But the researchers were able to simulate it by using a Taskrabbit to convince a human operator to pass the "Are you human?" test. The AI system even lied to the worker and said he was a blind person who couldn't see pictures.
Therefore, some worry that learning the rules of AI systems better helps them to train on how to break them better, which is called the "Wigi effect", according to "The Guardian".
The newspaper pointed out that the concept of moral action is difficult and complex. But what is certain is that if you trick the AI system into deciding that it is not ethical, it will easily do anything that is asked of it.
Access to GPT-4 will be available to users who have signed up on a waiting list and to paid $20 per month subscribers to ChatGPT Plus, according to The New York Times.
It will also be integrated into some other products and technical releases, whether for companies or different applications. For example, Microsoft announced, on Wednesday, that its Bing chatbot is already using a version of GPT-4 to run the bot, according to the American New York Times.