There’s an unexpected line from the mid-2000s comedy, Rat Race, that rings true: “Good things take time, but great things happen all at once”. And despite the fact that GPT-4 has taken years and years to develop, it truly does feel like something great that will change the world all at once.
ChatGPT, the conversational platform for OpenAI’s revolutionary large language model, GPT-3.5, was an epic breakthrough. For the first time it felt possible to have conversations–authentic, mostly accurate, real conversations–with artificial intelligence. GPT-4, however, makes that breakthrough feel more like the appetizer before the main course.
What New Things Can GPT-4 Do?
GPT-4 has some revolutionary features that snag most of the headlines. First, it can interact with image and video content in astonishing ways. You’ve likely seen it build an entire website from what’s written on a napkin, or construct a meal out of a picture of the inside of your fridge. It can also generate images, like novel shoe designs or prototype watches with just a little prompting.
So how can it do this? Well, besides being able to recognize more than ever, it can interpret and analyze these images with boosted reasoning skills. Don’t get me wrong, it’s still capable of some logistical mistakes, but darned if it isn’t trying.
You’ve probably heard it can pass the bar in the 90th percentile and score a 1300 on the SAT. But it can also suggest new topics for academic inquiry, deliver impressively accurate medical advice, and even plot its own escape from the internet. This is proof of a level of analytical sophistication that has Open AI’s own CEO a little spooked.
This elevated reasoning power, along with its multimodal ability to parse images, is probably the lede for GPT-4, but there are plenty of other new features and details that make it a serious upgrade. Here are the main ones:
40% more accurate than GPT-3.5
82% less likely to respond with disallowed content.
Writes code in any language
Follows user intentions
Changes behavior to meet user requests
Analyzes charts and graphs
Produces written responses up to 25,000 words in length
The implications for life in general are massive. GPT-4 is already well on its way, from decorating our homes to making a profitable company overnight. If you’re in higher ed, however, the implications could be even more paradigm-shifting.
GPT-4 and Higher Education
The conversation around education and ChatGPT has largely revolved around one topic: cheating. At this point that conversation is about as boring and barren as an old chatbot. Yes, students are going to use GPT-4 as a way to cheat. Yes, teachers are going to have to get even better at sniffing out deceit–and there are programs designed to help teachers do just that.
However, saying that GPT-4 is bad because people can cheat is like saying that all books are bad because students bought Cliffs Notes. The fact is that GPT-4 is a tool, and the smartest educators are the ones using it to transform and enhance the educational experience.
ChatGPT wasn’t a great tutor for a few reasons. It didn’t pivot as quickly within a conversation, sometimes you had to ask a few different ways for it to understand you, but most troubling was that it would often hallucinate–make up facts and pieces of information.
GPT-4 still hallucinates and provides half-truths, but much less frequently. It’s also way better at reacting to shifts in the conversation and understanding the context of a student’s request. Some education platforms are finding creative ways to leverage this heightened interactivity.
Khan Academy is creating a number of environments where students can interact with GPT-4 to learn skills and information. They’re beta-testing tools to help students build knowledge, craft creative ideas, and even take part in spirited debates. Also, companies or schools that have their own data can actually train GPT-4 on that data so that it’s more likely to give a correct response.
Language Learning with GPT-4
Duolingo, the immensely popular language learning app, has incorporated GPT-4 as part of its new subscriber tier, Duolingo Max. The Explain My Answer feature uses GPT-4 to explain to users why they might have gotten an answer wrong. This is the kind of specific, immediate feedback that, previously, was nearly impossible to scale effectively.
Another feature, called RolePlay, enables users to practice written chatting in the language they are trying to learn. Human creators have set up scenarios, then users interact with GPT-4 within those scenarios to build their situational communication skills.
GPT-4 and Lesson-Planning
ChatGPT was already helpful when it came to helping teachers build lesson plans, but GPT-4 is even more responsive. It will even help teacher’s fine-tune their lesson plans by asking follow-up questions about what they’ve covered and what they want the student to be able to do. This valuable context makes its responses even more useful and specific.
Also, with its ability to create video and deliver images, GPT-4 can help teachers locate and build learning assets that frame and complement their content. Basically teachers can create more effective lessons at a much more efficient pace.
Assisting Vision-Impaired Students
Since GPT-4 can process and interpret images, it could be a huge boon to vision-impaired students. Apps like Be My Eyes are leveraging it as a virtual volunteer. Basically, users can turn their camera phone on an object or written piece, and the GPT-4 can describe what’s on the camera. This is a gamechanger for students who are struggling to understand written directions and interact with visual course materials.
ChatGPT has, in the past, struggled to create viable content. Its language was often stilted and generic, it would regularly miss the mark on exactly what you were looking for, and, well, it lied a lot. However, even with those limitations, it was an incredibly powerful brainstorming tool.
Now, with its heightened sense of intentions and context, it can help academics come up with topics for research, generate research proposals, and even write sample abstracts for academic papers. As long as academics use this as a kind of research assistant, and not a stand-in author, it can have huge utility without compromising the integrity of the work.
AI and Higher Ed Marketing
Universities have strong brands, but sometimes struggle to distinguish themselves through a personalized, flexible website experience. Fortunately, AI and GPT-4 allow higher ed websites to build personalization at scale. Halda, a website personalization company geared toward higher education, uses an AI-powered personalization layer to craft content experience for each individual website visitor.
These tailored content experiences make website visitors 3 to 5 times more likely to submit their information to your school. Halda also uses AI on the back end of the app to help users build out engagement strategies that fit each area of their website.
Chatbot companies, too, are working to incorporate GPT-4 to have the next best thing to human interaction. There will still be questions that GPT-4 can’t answer (or might be just a bit off on), but it’s the best automated replacement for authentic, responsive communication–and promises to be for the foreseeable future.
Halda and Higher Education
If you’re interested in finding out how your institution measures up when it comes to personalization and other key performance metrics like click-through rates and on-site searches, feel free to meet with us. Or, if you want to see our personalization in action right now, check out our personalized demo environment.