(CNN) On the first day after its unveiling, GPT-4 surprised many users during the first tests and a demonstration of the company with its ability to write lawsuits, pass standardized exams and create a working website from a hand-drawn sketch.
On Tuesday, OpenAI announced the next-generation version of the artificial intelligence technology that powers its viral chatbot tool, ChatGPT. The more powerful GPT-4 promises to blow previous iterations out of the water, potentially changing the way we use the internet to work, play, and create. But it could also add to difficult questions about how AI tools can disrupt professions, enable students to cheat, and alter our relationship with technology.
GPT-4 is an updated version of the company’s Big Language Model, which is trained on large amounts of online data to generate responses to user prompts. It’s now available through a waitlist and has already made its way into some third-party products, including Microsoft’s new AI-powered Bing search. engine. Some users with early access to the tool share their experiences and highlight some of its most compelling use cases.
Here is an overview of the potential of GPT-4:
Analyze more than text
Basically, the biggest change to GPT-4 is its ability to work with photos that users upload.
One of the most mind-blowing use cases to date came from an OpenAI video demo that showed how a the drawing could be turned into a working website in a few minutes. The demonstrator uploaded the image in GPT-4, then pasted the resulting code into a preview that showed what a working website might look like.
In its announcement, OpenAI also showed how GPT-4 was asked to explain a joke from a series of images – which featured a smartphone with the wrong charger – and described why it was funny. Although it may seem simple, dissecting a joke is more complicated for AI tools to understand because of the necessary context.
In another test, The New York Times showed GPT-4 an image of the inside of a refrigerator and prompted him to come up with a meal based on the ingredients.
The photos feature is not yet available, but OpenAI should roll it out in the coming weeks.
Even easier coding
Some early adopters of GPT-4 with very little or no prior coding knowledge also used it to recreate iconic games such as Pong, Tetris or Snake after following the step-by-step instructions provided by the tool on how to proceed. Others made theirs original games. (GPT-4 can write code in all major programming languages, according to OpenAI.)
“The powerful language capabilities of GPT-4 will be used for everything from storyboarding to character creation to game content creation,” said Arun Chandrasekaran, analyst at Gartner Research. “This could spawn more independent game providers in the future. But beyond the game itself, GPT-4 and similar models can be used to create marketing content around game previews, generate news articles and even moderate game discussion forums.”
Similar to gaming, GPT-4 could change the way people develop apps. A user on Twitter said he makes a simple drawing app within minutes, while another claimed to have coded an app that recommends five new movies daily, plus providing trailers and details on where to watch them.
“Coding is like learning to drive – as long as the beginner gets guidance, anyone can code,” said Lian Jye Su, an analyst at ABI Research. “AI can be a good teacher.”
Pass the tests with flying colors
Although OpenAI said the update is “less capable” than humans in many real-world scenarios, it exhibits “human-level performance” on various professional and academic tests. The company said GPT-4 recently passed a mock law school bar exam with a score around 10% of top candidates. In contrast, the previous version, GPT-3.5, scored around 10%. The last version also scored well on LSAT, GRE, SAT, and many AP exams, according to OpenAI.
In January, ChatGPT made headlines for its ability to pass prestigious college-level exams, like that of the Wharton School of Business at the University of Pennsylvania, but not with particularly high marks. The company said it spent months using lessons from its test program and ChatGPT to improve the system’s accuracy and ability to stay on topic.
Provide more specific answers
Compared to the previous version, GPT-4 is able to produce longer, more detailed and more reliable written responses, according to the company.
The latest version can now give answers up to 25,000 words, down from about 4,000 previously, and can provide step-by-step instructions for the most unique scenarios, ranging from how to clean a piranha’s aquarium to extraction of DNA from a strawberry. An early user said he provided detailed pick-up line suggestions based on a question listed on a dating profile.
Rationalization of work in various industries
Joshua Browder, CEO of legal services chatbot DoNotPay, said his company is already work on the use of the tool to generate “one-click chases” to pursue bot callers, in an early indication of GPT-4’s vast potential to change the way people work across industries.
“Imagine receiving a call, clicking a button, [the] the appeal is transcribed and a 1,000-word lawsuit is generated. GPT-3.5 wasn’t good enough, but GPT-4 handles the job extremely well,” Browder tweeted.
Meanwhile, Jake Kozloski, CEO of dating site Keeper, said his company is use the tool to better match its users.
According to ABI Research’s Su, we may also see major breakthroughs in the “connected car” arena. [dashboards]remote diagnostics in healthcare and other AI applications that were previously not possible. »
A work in progress
Although the company has made vast improvements to its AI model, GPT-4 has similar limitations to previous versions. OpenAI said the technology has no knowledge of events that occurred prior to the cutoff of its dataset (September 2021) and does not learn from its experience. It may also make “simple errors in reasoning” or be “too gullible in accepting obvious misrepresentations from a user”, and fail to double-check work, the company said.
Gartner’s Chandrasekaran said this also reflects many AI patterns today. “Let’s not forget that these AI models are not perfect,” Chandrasekaran said. “They may produce inaccurate information from time to time and may be black box in nature.”
For now, OpenAI said GPT-4 users should exercise caution and exercise “high caution,” especially “in high-stakes settings.”