Tackling the 40-Page Manual with the Help of AI ‘Thesis Guru’ – Moss Shukla
We all have them: the long, detailed course manual, thesis manuals, style manuals. Manuals, manuals, manuals… Is there a manual for the manuals? We all know that they are critical for success, but that we also know that almost no one reads them from start to finish. For Moss Shukla, the bachelor thesis coordinator form the Psychology Department, this was a familiar problem. New supervisors, overwhelmed by a 40-page PDF guide, were often unaware of key rules and procedures.
His solution? He created a custom AI assistant, the BT Thesis Guru, to act as a friendly expert on the manual. His pilot project is a brilliant showcase of how AI Personas can make complex information accessible and save everyone valuable time.
Creating an Expert in 30 Minutes
Using the UvA AI Chat’s Persona feature, Moss uploaded the entire thesis manual PDF, gave the AI a name, and provided a simple instruction: answer any questions asked, but base the answers only on the content of the uploaded document. He also instructed it to cite the page numbers in the manual, so users could check the source and find the detailed answer.
The result was a powerful new tool, and the setup was shockingly fast. “I mean, it was trivial,” Moss noted, estimating it took him “maybe half an hour at most” to get a working version up and running. Supervisors could now simply ask the Guru a question like “How many meetings should I have with the student and what should they be about?” and get a synthesized answer drawn, if needed, from multiple sections of the manual at once. The idea was met with universal enthusiasm from the other coordinators he showed it to. And now that the UvA AI Chat is available to all teachers and staff, he will be able to share it with all the supervisors. .
A Word of Caution: The Szechuan Recipe Test
While the tool is incredibly effective, Moss offers a crucial piece of wisdom for anyone using generative AI. He calls it the Szechuan dish test.
“Let’s say that I want to learn about a recipe for a Szechuan dish,” he explains. “I ask for an excellent recipe, and it gives me one. It looks like a good recipe to me, so I make it, and the dish turns out to be just complete ****. And the reason is because you cannot evaluate yourself whether this is a good recipe or not. Not until you make the food and taste it”.
The AI output almost always looks plausible and correct. But if you aren’t an expert chef, you can’t spot the flaws in the recipe until you’ve already served the meal. The same is true in education. “It’s working well for you because you already are an expert,” Moss points out. A student who doesn’t understand statistics, for example, cannot judge whether the AI’s proposed solution is brilliant or completely wrong.
The most important takeaway from this pilot is that AI tools are most powerful in the hands of someone who has the expertise to critically evaluate the output. The tool can help you write code, draft an email, or summarize a document, but you need to “understand what it did” to know if the result is any good.
For teachers looking to replicate Moss’ success, the message is twofold:
- Do try this at home! Creating a persona to make a dense document accessible is easy, fast, and incredibly useful.
- Remember the Szechuan recipe. Use AI to augment your own knowledge and save time, but never trust the output blindly. The ability to critically judge the AI’s work is the most important skill of all.