We don’t often like putting ourselves in the spotlight, but sometimes it’s worth doing precisely that – especially when you can do it safely in the comfort of your home.

Don’t use GenAI as an answer engine.

Use it as a question engine.

There’s now solid research showing that testing is an effective strategy for enhancing learning and retention compared to restudying alone. What you want, however, is instant constructive feedback on the answers that highlights the gaps in your knowledge but also helps fill them.

Guess what’s good at doing that? Frontier large language models, i.e. chiefly OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini (and a few others).

Instead of asking them for information, ask them to test you on a topic. You can use this prompt to get started:

Please test my knowledge on a topic of my choice with at least 20 questions, gradually increasing in difficulty. Start by asking me which topic I want to be tested on. The questions should cover various subtopics and include a mix of formats (e.g., multiple-choice, short-answer, open-ended), tailored to the topic.

After I submit each answer, provide a brief evaluation of my accuracy (e.g., ‘Well done, that’s mostly correct’ or ‘Not quite’), followed by the correct answer and an in-depth, analytical explanation. Progress to harder questions more quickly if my initial responses indicate a solid grasp of the topic. No hints or skipped questions. The test is self-paced, with no time limit.



Hint: for a classic and instant demonstration of the phenomenon of illusion of explanatory depth, ask it to test you on bicycle mechanics, and then realize you – at least if you’re anything like me – will fail at answering questions that really seem quite simple

Another tool that you might want to explore is globe.engineer, specifically the question mode of index.globe.engineer which will take you through a guided learning exercise, starting by querying your knowledge about the topic and providing instant feedback on how you’re doing:

The frontier LLMs are broadly speaking great at this. Use them for testing yourself.

Leave a Comment

Your email address will not be published. Required fields are marked *