Remember in 2022 when traditional graded essays seemed to become obsolete overnight? ChatGPT’s introduction to the world of education was not the most positive experience for most. The plagiarism issue hasn’t gone away, but it feels like a much smaller deal now. The exponential growth of generative AI has introduced new innovations, new opportunities, and plenty of new challenges. Let’s take a look at what else has happened in just over one year of generative AI evolution in schools.
The Technical Improvements
GPT updates
Although many new large language models have been released in the past year, OpenAI’s ChatGPT has become the de facto standard for Gen AI. One year ago, the program was based on a version known as GPT-3.5. Since March 2023, paid users (and the many applications that leverage ChatGPT) are now utilizing GPT-4. This latest version has been called “10 times more advanced” than its predecessor. Here are just a few of the improvements:
- Where GPT-3.5 was a text-only model, GPT-4 can analyze and produce text, images, and voice.
- GPT-4 is far more capable of understanding nuance and context than previous models. All indications are that its training parameters are “an order of magnitude” greater than GPT-3. When GPT-3 took the bar exam, it finished in the 10th percentile. GPT-4 hit the 90th percentile. Similar results occurred with other tests, including the Biology Olympiad (31st percentile vs. 99th).
- Where GPT-3.5 was limited only to information through late 2021, GPT-4 can access the internet to stay up to date on even the most recent developments in the world.
- GPT-4 is safer and more accurate. Per OpenAI, it is “82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5.”
ChatGPT, like all the other LLMs that have and will come after, is growing up before our eyes. It’s just doing so at a much faster rate than many anticipated. For those who jumped on the bandwagon early, the conversational difference today compared to a year ago is remarkable.
Image enhancements
AI image generation has evolved in parallel with text generation models. In the past year, the output has gone from novel and quirky to largely indistinguishable from “real” images and art. OpenAI released DALLᐧE 3 in October 2023, Midjourney evolved from V4 to V6 over the course of the year, and Stable Diffusion became about 3.5x larger and far more powerful in July 2023 with its XL 1.0 release.
Some of the most obvious impacts of AI evolution in image generation include:
- The ability to more accurately represent human anatomy. AI image generators struggled mightily with hands, ears, and other quirky body parts in its early iterations, but no more.
- Improved text handling. There are still some issues here, but compared to where we were a year ago, AI image generators do a much better job of including properly spelled text in images where appropriate. This is in stark contrast to the nonsense words and imaginary letters that were all too common a year ago.
- Improved content moderation. While most of the major players are struggling to find the right balance between censorship and artistic freedom, built-in content moderation has significantly reduced the ability to play fast and loose with public figures and inappropriate content.
- Bias in training data persists, but the past year has at least brought more awareness to the issue and at least some steps in the right direction. OpenAI has been among the most vocal about pledging to address “harmful biases related to visual over/under-representation.” In general, AI image generators still fall back on stereotypes more often than not.
The lines continue to blur between machine and art, and the improved performance is a double-edged sword. While image outputs are most certainly becoming more realistic and lifelike, they are also becoming harder to identify as AI-generated. The inability to distinguish what’s real from what’s not is one of the biggest challenges we face going forward.
The Biggest Wins in K-12
Teacher time-savers
Administrators have their work cut out for them to raise awareness of how AI can help teachers with tedious, non-instructional tasks. But one thing we’ve learned this year is that the most tech savvy educators are already getting it done. In EdTech Evolved’s end-of-year survey of district tech leaders, 61% said their teachers were already using AI. 48% said they have already seen positive results from that usage.
Any time something comes along that makes the teaching profession a little less demanding, we have real cause for celebration. Generative AI is taking administrative tasks off of teachers’ plates and empowering them to focus more on their instruction and their students.
Teachers are going directly to the source (ChatGPT), while also exploring apps that tailor the Gen AI to specific needs. The most popular teacher time-saving use cases for Gen AI include:
- Lesson plans
- Quiz creation
- Presentation builders
- All of these recommendations (with examples) from OpenAI
Deeper differentiation and personalization
The Gen AI evolution has been slower to touch on the student experience, but it is happening on two fronts. First, a fair number of students are taking the ball and running with it on their own. That came to light immediately with the use of ChatGPT for essay writing, and students are only becoming more savvy and finding more tools with time. Second, both established edtech providers and new startups have been rolling out Gen AI features or product offerings since before the start of the school year.
In the aforementioned survey, 44% of district tech leads said the aspect of AI they were most excited about was “the ability to personalize learning based on student levels, needs, and interests.” From 1:1 AI tutors to playfully personalized decodable readers, this innovation is already happening before our eyes. Once districts have some time to get their feet under them and establish stronger vetting processes, students will reap the benefits of this kind of next-level personalization on a much larger scale.
The Biggest Challenges
The darker side of Gen AI
We all know about the cheating problem with ChatGPT. But ongoing AI evolution has introduced even more variables to the conversation. In the past year, we’ve seen students busted for using the technology for everything from book reports to high-stakes testing. We also saw some far more nefarious applications of Gen AI that will pose heightened challenges for the entire school community, from students and parents to faculty and administration.
The threat of deepfakes looms large in the coming year. In at least one New Jersey high school, students used Gen AI to create and distribute pornographic images of their classmates. The same technology that can support higher levels of student engagement and expand the possibilities of digital exploration will also make it harder for us to distinguish facts from misinformation in every aspect of our lives.
With a polarizing election year looming and society’s growing reliance on social media and digital communications, the potential for fraud, social engineering, and other malicious applications of Gen AI is something everyone should be concerned about.
Keeping pace with change
Educators have always been playing catchup with technology. They’ve spent years fighting for more accessible and equitable broadband access. Then they took on the challenge of keeping up with device upgrades. More recently, they’ve shouldered the burden of navigating an increasingly complex web of data privacy and security legislation. Gen AI is no different.
As of the end of 2023, only 4% of surveyed district technology leaders reported having a formal, documented policy governing the use of AI. Another 39% said they were working on it. That left more than 65% feeling like they were behind the curve in addressing the use of AI. There is some question about how quickly the technology will continue to evolve (has it plateaued already?), but the hope is that most districts will be able to at least find their footing before the start of next school year.
What Happens Next for AI Evolution?
Whether intentional or not, AI evolution seems to have plateaued, or at least slowed, compared to the exponential growth many predicted at the start of 2023. GPT-5 is “on the roadmap” for OpenAI, but its release does not appear to be imminent. The most recent updates to Gen AI models in general have been more incremental than game-changing.
This respite, however brief it may be, might be exactly the window of opportunity we need to get caught up from a policy and preparedness standpoint. For school districts in particular, the next few months should offer enough time to integrate AI into acceptable use policies, purchasing processes, and technology applications. The resources to do so already exist, now it’s just a matter of assigning accountability and getting things done.
If you’re asking yourself whether “this is it” as far as AI evolution goes, the answer is assuredly no. Generative AI will continue to have a generational impact on productivity, creativity, and the workforce of the future for many years to come. For now, we should all take a deep breath, take stock of how we can use this technology in our own lives, and educate, educate, educate whenever possible.
Keep your finger on the pulse of the AI evolution and its impact on the K-12 community by subscribing to EdTech Evolved.
