At NASA, innovation is the driving force behind everything from Apollo to Artemis. Testing new concepts is a key part of NASA’s innovation, and it even applies to teaching and training. When the Open Science Training team aimed to create a streamlined version of NASA’s Open Science 101 (OS101) course, Open Science Essentials (OSE), they faced a challenge: a tight timeline and a mountain of content to refine. Instead of tackling this task the traditional way with manual review, writing, and formatting, they decided to try a new approach. They turned to large language models (LLMs) to help speed up the work.

This collaboration between human experts and artificial intelligence was more than just an experiment. It was an example of how thoughtful implementation of new tools can help us reach our goals faster and more efficiently. By strategically using LLMs, the team was able to transform a lengthy slide deck into a comprehensive, interactive course in just two months, significantly reducing development time.

Accelerating Development with AI

The primary hurdle for the team was converting their course content from slides into a structured, online format known as MDX. This process, if done manually, would have been a time-consuming endeavor involving multiple rounds of coordinated review. Developers would have had to transcribe text, ensure consistency, and integrate specific code components. Instead, the OSE development team opted to leverage an LLM to assist with these initial steps.

By prompting an LLM with clear instructions and using additional AI tools to refine output, the team was able to automate the conversion process. The LLM generated output from the slides in the desired format, effectively eliminating the need for manual transcription and conversion. This automation saved the team considerable time, freeing them up to refine the material instead of getting bogged down in formatting.

Image

This novel approach to course development was enabled by internal collaboration with the LLM team within NASA’s Data and Analysis Services Project (DASP). Brian Ressler, an Open Science Training team developer supporting the DASP team on contract with Media Fusion, worked together with Carson Davis, a researcher supporting DASP at the University of Alabama in Huntsville (UAH), to determine the most effective prompts and models to deploy. Once Ressler and Davis optimized their approach, the LLM generated the desired MDX output quickly, enabling the OSE development team to forge ahead with refining the content.

Collaborative Course Creation

A significant benefit of using a LLM was how it empowered the entire development team, including those without a technical background. Often, creating or modifying an online course layout is a task reserved for web developers. However, leveraging an LLM allowed non-technical team members to more actively participate in the development process.

LLMs were used to generate a draft course layout based on OS101 PowerPoint content. If a suggested webpage structure needed edits, the team members only needed to learn how to read the documentation for the course's code components. This documentation was also generated by an LLM, so creating new sections or pages simply involved copying, pasting, and replacing content within a pre-defined framework. This approach not only accelerated the development timeline but also made the editing process more creative and collaborative, encouraging additional voices and perspectives to shape the final product.

The Critical Role of Human Oversight

LLMs proved to be a valuable tool in OSE course development, but it's crucial to emphasize that LLMs are just that – tools. They served as a powerful assistant, but the final authority and creative direction were managed by the human team. Throughout the entire process, OSE development team members at both NASA’s Marshall Space Flight Center in Huntsville, Alabama and NASA’s Ames Research Center in Mountain View, California, were actively involved in reviewing, editing, and refining the generated content.

LLMs also helped generate supplementary content to fill in gaps and enrich course material, but it was the human team that provided oversight and ensured the scientific integrity and presentation of the course. The team also used an LLM to extract images from the OS101 slide deck, saving them the manual effort of searching for and downloading each file individually.

This human-in-the-loop approach was essential to successful curriculum development. It ensured that the course was not just technically sound but also engaging, accurate, and aligned with supporting NASA’s open science standards. 

The OSE development workflow shows how AI can amplify human creativity and productivity while also helping us reach our goals faster than before. It's a model of collaboration that serves as an example for future projects, demonstrating how NASA’s spirit of exploration and innovation can inspire all that we pursue.

Funded by NASA’s Office of the Chief Science Data Officer, the Open Science Training team is part of DASP, which operates within Marshall’s Office of Data Science and Informatics (ODSI).