On the first day it was shown to the public, GPT-4 surprised many users with its ability to write lawsuits, pass standardized tests, and build a website from a hand-drawn sketch.
On Tuesday, OpenAI announced the next-generation version of the AI technology that powers ChatGPT, its popular chatbot tool. The more powerful GPT-4 promises to blow the previous versions out of the water. This could change how we work, play, and make things on the internet. But it could also raise more complex questions about how AI tools can change jobs, let students cheat, and change how we use technology.
GPT-4 is a newer version of the company’s large language model, which uses a lot of online data to learn how to give complex answers to user questions. It is now available through a waitlist, and some third-party products, like Microsoft’s new AI-powered Bing search engine, are already using it. Some early users are talking about their experiences with the tool and pointing out some of the most exciting ways it can be used.
Here’s a closer look at what GPT-4 might be able to do:
More than just reading the text
The most crucial change to GPT-4 is that users can now upload and work with photos.
One of the most impressive uses of AI so far was shown in an OpenAI video demo, which showed how a drawing could be turned into a working website in just a few minutes. The demonstrator uploaded the picture to GPT-4 and then pasted the resulting code into a preview that showed how it could be a functional website.
In its announcement, OpenAI also showed how GPT-4 was asked to explain a joke from a series of images, which included a smartphone with the wrong charger, and why it was funny. Even though it might sound simple, it’s harder for artificial intelligence tools to figure out what a joke is about because they need more context.
In another test, The New York Times showed GPT-4 a picture of the inside of a refrigerator and asked it to make a meal from what it saw.
The photos feature has yet to run, but OpenAI plans to do so in the next few weeks.
Coding Made Even Easier
Some early GPT-4 users with little or no coding experience have also used it to make classic games like Pong, Tetris, or Snake by following the tool’s step-by-step instructions. Some people have made their games. OpenAI says that GPT-4 can write code in all major programming languages.
Arun Chandrasekaran, an analyst at Gartner Research, said, “The powerful language skills of GPT-4 will be used for everything from storyboarding to making characters to making game content.” “In the future, this could lead to more independent game providers. But GPT-4 and similar models can be used for more than just the game itself. They can be used to create marketing content for game previews, write news articles, and even moderate gaming discussion boards.
GPT-4 could change how apps are made, similar to how games change. One Twitter user said they earned a simple drawing app in minutes, and another said they coded an app that recommends five new movies every day, along with trailers and information about where to watch them.
Analyst at ABI Research, Lian Jye Su, said, “Learning to code is like learning to drive. Anyone can learn to code if a beginner has someone to show them the ropes.” “AI can be an effective teacher.”
Getting perfect scores on tests
OpenAI said the update is “less capable” than humans in many real-world situations, but it shows “human-level performance” on several professional and academic tests. The company said that GPT-4 recently passed a simulated law school bar exam with a score in the top 10% of all test takers. The old version, GPT-3.5, on the other hand, scored in the bottom 10%. OpenAI says the latest version did well on the LSAT, GRE, SATs, and AP exams.
In January, ChatGPT made headlines because it could pass prestigious graduate-level exams, like one from the Wharton School of Business at the University of Pennsylvania, but not with very high scores. The company said it spent months using what it learned from its testing program and ChatGPT to improve the system’s accuracy and ability to stay on topic.
Providing More Precise Responses
The company says written responses from GPT-4 are longer, more detailed, and more reliable than those from the previous version.
The new version can now give answers of up to 25,000 words, up from about 4,000 before. It can also provide detailed instructions for even the most unusual situations, like cleaning a piranha’s fish tank or getting the DNA out of a strawberry. One of the first people to use it said it gave detailed ideas for pickup lines based on a question on a dating profile.
Streamlining Work in Different Fields
Joshua Browder, CEO of the legal services chatbot DoNotPay, said that his company is already working on using the tool to generate “one-click lawsuits” to sue robocalls. This is an early sign of how GPT-4 could change how people work in many different industries.
“Imagine getting a phone call, clicking a button, and getting a 1,000-word lawsuit because of it. “GPT-3.5 wasn’t good enough, but GPT-4 does a great job,” Browder wrote on Twitter.
The CEO of the dating site Keeper, Jake Kozloski, said that his company is using the tool to find better user matches.
Su at ABI Research says that it’s possible that “connected car dashboards, remote diagnosis in healthcare, and other AI applications that weren’t possible before” will also make significant steps forward.
Work in Progress
Even though the company has made many changes to its AI model, GPT-4 still has the same problems as earlier versions. OpenAI said that the technology doesn’t know what happened before September 2021, when its data set ends, and it needs to learn from what it does. The company said it could also make “simple reasoning errors” or be “too trusting of obvious false statements from a user” and not double-check its work.
Chandrasekaran of Gartner said this is true of many AI models today. Chandrasekaran said, “Let’s not forget that these AI models aren’t perfect.” “They can sometimes give out wrong information and be a black box.”
OpenAI said that for now, users of GPT-4 should be careful and use “great care,” especially “in high-stakes contexts.”
You must log in to post a comment.