(CNN) On the first day after opening. GPT-4 surprised many users in the first tests, and the company demonstrated the ability to prepare cases, pass standardized tests, and build a functional website from a hand-drawn design.
On Tuesday, OpenAI announced The next-generation version of artificial intelligence technology that powers the viral chatbot tool called ChatGPT. More powerful GPT-4 promises Blowing previous iterations out of the water, it could change the way we use the internet to work, play and create. But it also raises challenging questions about how AI tools can enhance careers, help students cheat, and improve our relationship with technology.
GPT-4 is an improved version of the company’s large language model, which is trained on massive amounts of online data to generate complex data. Answers to user questions. It’s now available through a waiting list 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 are sharing their experiences and highlighting some of its most compelling use cases.
Take a closer look at GPT-4’s capabilities:
Analyzing more than text
At its core, the biggest change to GPT-4 is the ability to work with photos that users upload.
Here comes one of the most jaw-dropping use cases so far OpenAI video demo How a A picture can be turned into a functional website. In minutes. The demo uploads the image to GPT-4 and pastes the resulting code into a preview that shows how a working web page might look.
In the announcement, OpenAI also He showed. GPT-4 explained how he was asked to explain a joke from a series of images – one showing a smartphone with a faulty charger – and why it was funny. Although it seems straightforward, the division of humor It is more complicated to adopt artificial intelligence tools because of the important context.
In another test, The New York Times He showed. GPT-4 is an image of the interior of the refrigerator and prompts food to arrive based on the contents.
The Photos feature hasn’t been released yet, but OpenAI is expected to unveil it in the coming weeks.
Coding just got easier.
Some early users of GPT-4 used it with little or no prior coding knowledge. Recreate popular games like Pong, Tetris or Snake after following the step-by-step instructions provided by the tool. Others have done their own Original games. (GPT-4 can be coded in all major programming languages according to OpenAI.)
“GPT-4’s powerful language capabilities are used for everything from storyboarding, character creation to game content creation,” said Arun Chandrasekaran, an analyst at Gartner Research. “This may create 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 host game discussion boards.”
Similar to gaming, GPT-4 can change the way people develop applications. said one user on Twitter. Made a simple drawing application In minutes, he said he had another Coded An application that recommends five new movies Along with providing daily movie trailers and details on where to watch.
“Coding is learning how to drive – anyone can code as long as the beginner gets some guidance,” said ABI Research analyst Lian Jye Su. “AI can be a good teacher.”
Pass the tests with flying colors
Although OpenAI he said. The update is “less capable” than humans in many real-world situations, showing “human-level performance” on various vocational and academic tests. The company says the GPT-4 recently passed the law school bar exam with the top 10% of test takers. In contrast, the previous version, GPT-3.5, scored around the bottom 10 percent. Latest version It performed strongly on the LSAT, GRE, SATs, and many AP exams, according to OpenAI.
In January, ChatGPT made headlines for its capabilities. Pass the prestigious graduate examsLike one from Pennsylvania’s Wharton School of Business, but not particularly high-scoring. The company says it spent months using lessons learned from its pilot program and Chat GPT to improve its system’s accuracy and ability to stay on topic.
Giving more accurate answers
Compared to the previous version, GPT-4 can generate longer, more detailed and more reliable text responses, the company said.
The new version now offers answers of up to 25,000 words, up from 4,000 before, and can provide detailed instructions for even the most specialized situations, from how to clean a piranha fish tank to extracting strawberry DNA. One early user said he made deep suggestions for pick-up lines based on a question listed on a dating profile.
Facilitate operations in various industries
Joshua Browder, CEO of legal services chatbot Donotpay, says his company already is. Working on using the tool To generate “one-click lawsuits” to prosecute robocallers, GPT-4 is already showing its vast potential to change how people work across industries.
“Imagine receiving a call with the click of a button. [the] A call was transcribed and a 1,000 word lawsuit was filed. “GPT-3.5 wasn’t good enough, but GPT-4 is up to the task,” Browder tweeted.
Meanwhile, dating site Guardian CEO Jake Kozloski says his company. Using the tool It better relates to the users.
According to ABI research, we can also see major developments in the “connected car [dashboards]remote diagnostics in healthcare and other previously impossible AI applications.
A work in progress
Although the company has made extensive improvements to the AI model, GPT-4 has the same limitations as the previous versions. OpenAI technology is available It has no knowledge of events that occurred before the data set was terminated (September 2021) and does not learn from experience. It can also make “simple errors of judgment” or be “overzealous in accepting clearly false statements from a user” and cannot double-check work, the company said.
Gartner’s Chandrasekaran says this also reflects many AI models today. “We must not forget that these AI models are not perfect,” said Chandrasekaran. “From time to time they can produce inaccurate information and become a black box in nature.”
For now, OpenAI says GPT-4 users should exercise caution and use “great caution,” especially in “high-level contexts.”