CNN
—
On the first day of the unveiling, With the ability to draft lawsuits, pass standard exams, and build working websites from hand-drawn sketches, GPT-4 wowed many users in early tests and corporate demos.
OpenAI on Tuesday unveiled the next-generation version of the artificial intelligence technology behind its viral chatbot tool ChatGPT.The Promise of a Stronger GPT-4 It has the potential to blow previous iterations out of the water and change the way we work, play and create using the internet. But it also has the potential to raise even more difficult questions about how AI tools can upend jobs, enable student cheating, and change our relationship with technology.
GPT-4 is an updated version of the company’s Large Scale Language Model, which is trained on vast amounts of online data to produce complex language models. Responses to user prompts. It’s available now via a waiting list and is already in some third-party products, such as Microsoft’s new AI-powered Bing search. engine. Some users who had early access to the tool shared their experiences and highlighted some of its most compelling use cases.
Let’s take a closer look at the possibilities of GPT-4.
The biggest change in GPT-4 is the ability to manipulate user-uploaded photos.
One of the most amazing use cases ever came out of the OpenAI video demo. You can turn your drawing into a functional website within minutes. The demonstrator uploaded a photo to her GPT-4 and pasted the resulting code into the preview to show it working as a website.
In its announcement, OpenAI also showed how GPT-4 was asked to explain a joke in a series of images (featuring a smartphone with the wrong charger), explaining why it was funny. . It may sound simple, but if you analyze the joke, It is more complex for artificial intelligence tools to perceive because of the required context.
In another test, The New York Times showed GPT-4 a picture of the inside of a refrigerator and encouraged it to come up with meals based on ingredients.
The photo feature hasn’t been rolled out yet, but OpenAI plans to roll it out in the coming weeks.
Some early GPT-4 users with little or no prior coding knowledge use GPT-4 to: Recreate iconic games like Pong, follow the instructions provided by the tool to run Tetris or Snake.others made their own original game. (OpenAI says GPT-4 can code in all major programming languages.)
“The powerful language capabilities of GPT-4 will be used for everything from storyboarding and character creation to game content creation,” said Gartner Research analyst Arun Chandrasekaran. . “This could lead to even more independent game providers in the future. But beyond the games themselves, GPT-4 and similar models will be used to create marketing content around game previews, news You can use it to write articles or even be a moderator for game discussion boards.”
Like games, GPT-4 could change the way people develop apps. A Twitter user said: I made a simple drawing app In a few minutes another coded An app that recommends 5 new movies Each day we will provide trailers and details of where you can watch them.
ABI Research analyst Lian Jye Su said, “Coding is like learning to drive, even beginners can code with guidance.” “AI can be a good teacher.”
OpenAI said the update was “less capable” than humans in many real-world scenarios, but showed “human-level performance” in various professional and academic tests. GPT-4 recently passed a law school mock bar exam with scores in the top 10% of test takers, according to the company. In contrast, his GPT-3.5 score on the previous version was in the bottom 10%.newest version According to OpenAI, they also performed well on the LSAT, GRE, SAT, and many AP exams.
In January, ChatGPT made headlines for its ability to pass prestigious graduate-level exams such as the University of Pennsylvania’s Wharton School of Business, but it didn’t do particularly well. The company said it spent months using its testing program and lessons learned from ChatGPT to improve the system’s accuracy and ability to stay on topic.
The company says GPT-4 can produce longer, more detailed, and more reliable written responses than previous versions.
The latest version can now provide responses of up to 25,000 words, up from about 4,000 previously, providing detailed instructions for even the most unique scenarios, from how to clean a piranha tank to extracting strawberry DNA. Now available. One early user said it provided detailed suggestions for pick-up lines based on questions in their dating profiles.
Joshua Browder, CEO of legal services chatbot DoNotPay, said his company already working on using tools This is an early indication of the great potential of GPT-4 to change the way people work across the industry, with the aim of filing a “one-click lawsuit” against robocallers.
“Take a call, click a button, [the] The phone call is transcribed and a 1,000 word lawsuit is generated. GPT-3.5 wasn’t good enough, but GPT-4 does the job very well,” Browder tweeted.
Meanwhile, Jake Kozloski, CEO of dating site Keeper, said his company using the tool Better matching with users.
According to ABI Research’s Su, we could also see big strides in the “connected car.” [dashboards], remote diagnostics in healthcare, and other AI applications not previously possible. ”
Although the company has made significant improvements to its AI models, GPT-4 has similar limitations as previous versions. OpenAI said the technology We lack knowledge of what happened before the data set died out (September 2021) and have not learned from that experience. It may also make “simple reasoning errors” or “excessively deceive users with blatantly false statements” and may not double-check their work, the company said. Stated.
Gartner’s Chandrasekaran said this also reflects many AI models today. “Remember, these AI models are not perfect,” Chandrasekaran said. “They can generate inaccurate information at times and can be essentially black boxes.”
For now, OpenAI said GPT-4 users should exercise caution and be “very careful”, especially in “high-stakes situations”.
