Red Hat's Brno office has been helping Masaryk University teach a technical writing course since 2022, after a pandemic-related delay. At a talk at the Open Source Summit in Prague, Red Hat's Jiří Herrman described what the course tutors learned from students in the Faculty of Informatics. The course, PV177, was designed for students with some technical background, and its materials are open source.
Several approaches failed. Assignments using shell-based tools did not work because some students had never used the command line, some had borrowed laptops without administrator access, and a few brought only phones. Peer grading, meant to mirror professional review, also went poorly, so the tutors taught students how to grade one another's work without causing offense. Remote lecturers did not work well either, especially when the tutor changed each session. What did work included awarding points on a semi-arbitrary basis with occasional secret prizes, delivering feedback as internet memes, and a final Dungeons & Dragons-style role-playing exercise where students navigated a cross-functional team to produce release notes and documentation.
The biggest challenge was LLM use. Students were told not to use LLMs, but many did anyway. Praising good work that used LLMs encouraged more cheating, while scolding discouraged students and made them cheat more. The course team therefore decided to teach students how to use LLMs properly instead of banning them. Because the course materials are open source, graduates also leave with a public portfolio of coursework they can show employers.
A related Birds of a Feather session at the same conference asked tech writers how they use AI. An overwhelming majority said they use it, but there was broad agreement that LLMs are poor at writing documentation themselves, at creating or reading diagrams, and at reviewing docs because they are too positive. Writers found LLMs useful for identifying what changed in a large product and needs new docs, documenting decisions from meeting notes, summarizing conference notes, reverse-engineering code to understand intent, and finding existing docs to avoid duplication.