Just when we thought artificial intelligence had hit its peak over the last few months, OpenAI, the creator of ChatGPT, announced on Tuesday a new generation of the chatbot that knocks it out of the park.
GPT-4, is an updated version of the company’s language model, which has been trained using significant amounts of online data to generate detailed responses to user questions and prompts. Though it’s only available via a waitlist, it’s already been incorporated into various third-party products like Microsoft’s new AI-powered Bing search engine.
Those lucky enough to be given early access have shared their experiences with it, and their experiments are pretty astounding.
For example, GPT-4 does more than generate impressive essays and answer math questions. In one instance, a user demonstrated how a drawing can be turned into a fully functioning website in minutes. In the video, the program takes a messy notebook drawing and transforms it into a lackluster, yet functioning website.
🤯 OpenAI's GPT-4 release is turning heads by turning a messy notebook drawing into a fully-functioning website. This is a game-changer in website building! pic.twitter.com/Szt5R8djre
— The Joker (@Neo19890) March 15, 2023
It also can help students dive into coding with little to no prior knowledge. In another demonstration, one user had it recreate the iconic game Pong in under 60 seconds. All they had to do was follow step-by-step instructions provided by the tool and boom, one of the most iconic games in history was recreated.
I don’t care that it’s not AGI, GPT-4 is an incredible and transformative technology.
I recreated the game of Pong in under 60 seconds.
It was my first try.
— Pietro Schirano (@skirano) March 14, 2023
As far as test-taking goes, it’s a star student. According to OpenAI’s announcement, “It passes a simulated bar exam with a score around the top 10% of test takers.”
Here are some other exams GPT-4 has taken:
- SAT Evidence-Based Reading and Writing: Scored 710/800 (93rd percentile)
- Medical Knowledge Self-Assessment Program: Scored a 75%
- AP US Government: Scored a 5 (88th-100th percentile)
So what does this mean for higher education? Well, if educators approach it the same way they did ChatGPT, we may see a divide between those who believe it should be embraced and those who fear it will promote cheating and academic dishonesty. Sal Khan, founder and CEO of Khan Academy, recently published a blog post describing his experience with GPT-4. Simply put, he thinks it has massive potential to close the digital divide among students.
For example, the organization has implemented a small AI pilot called Khanmigo for students and teachers to serve as a 24/7 tutor. Users can quite literally ask questions they would ask of their in-person tutor and it will generate patient, human-like responses. And for teachers, it serves as a timesaver so they can focus on what’s important—their students.
For example, in this video featured in Khan’s blog post, a user is seen interacting with AI to receive help with a math problem.
However, there’s still plenty of room for growth, according to Khan.
“AI makes mistakes,” he wrote. “Even the newest generation of AI can still make errors in math. AI can still ‘hallucinate,’ which is the term the industry uses for making stuff up. A lot of work needs to be done.”