Using AI! That said, we've hand-written thousands of challenges, and used those to seed these newly generated ones - it's not just one-shot slop. The generation loop is an agentic system that takes the following into account:

• Spaced repetition: We use a time-variable algorithm to surface topics you may need a refresher on.
• Personalization: We try to only give you challenges that require knowledge you've already learned.
• Mastery: We weight toward topics that you struggled with in the past.
• Correctness: Our system runs the challenges to ensure generated solutions work before you start.
• Variety: There are 3 types of challenges: write the code, fix the bug, and interview question.

If you don't enter a custom prompt, you'll get a challenge based solely on your learning history. If you do enter a custom prompt, we'll still use your history, but the system will override any of its choices with your instructions if they contradict.

Game mechanics like streaks, quests, sharpshooter etc. all work the same way in the Training Grounds as they do in the courses, but the base XP gain of challenges is 25% lower than in regular courses. Additionally, XP gain is reduced by 5% per challenge to a maximum of -50% per day.

For now, we're solely focused on generating high-quality challenges, which means we're not spending much time trying to optimize for speed. Speed improvements are coming in the future!

The reason it takes so long at the moment is that we're pulling a bunch of stuff about your history into context so we can generate a highly relevant challenge, and once generated, we run the challenge on our backend, then allow the AI to fix any issues before serving it to you.

Challenges are AI generated, so occasionally you'll get some weird stuff, even with all the safeguards, prompts, tests, and hand-crafted examples we've put in place. Be sure to use the star ratings and skip button to give us feedback so that we can keep improving the system.