Slideshows are one of the most popular ways teachers deliver lessons in the classroom.
Whether they use Google Slides, PowerPoint, Canva, or another tool, teachers put a lot into finding quality images, appropriate content, and curriculum-aligned information. That means first drafts can take a lot of time and attention. And all that time and attention might be better spent with students, rather than with technology.
That's why our AI team has been developing an AI-powered slide-generation tool. This prototype is fast and reliable, and best of all, it keeps the teacher in the loop with full control over the final lesson.
In our demo video, we start entering our topic of UV Rays. We then select our grade - in this case, grade 5. Then we enter our learning objective for the lesson – which is the effect of UV rays on the human body.
Then the AI does its thing. In a few seconds, the AI delivers a recommended outline for the lesson. We get five high-level topics, each with several supporting slides and suggested activities.
A critical feature of our slide-generation tool is that the teacher is always in full control. Everything from the outline, to the slide topics, to the number and order of slides can be edited. Additionally, users are able to add supporting slides, topics, and activities.
When teachers are happy with their outline, they can generate a first draft.
In recent months, we've seen tools like ChatGPT and its ability to generate copy, essays, and even code. But slide-generation takes generative AI a step further.
Among others, there are two big challenges we're tackling here.
For starters, how do you ensure curriculum alignment? Different districts have different learning objectives, so our slide-generation tool needs to align with those discreet objectives. That's where domain-specific Large language models (LLMs) come into play.
The second challenge is, how do you get relevant images? LLMs can generate images, but there are some quality concerns. So, our tool uses Bing image search to find the most relevant images.
Once the slideshow is created, teachers can review the attribution for the images the AI retrieved. Users also can review other recommended images and replace them as they see fit.
If teachers are not happy with the content, they also have the option to regenerate, and they can choose a new learning objective.
When the presentation is perfected, teachers can save as a PDF or PowerPoint, or they export to Google Slides.
Stay tuned as our slide generation prototype develops and for more last-mile solutions for education from Merlyn Mind's AI Labs team.