Hugo Tunius, a developer and hacker culture enthusiast, has developed an elegant workaround for one of modern computing's most pervasive problems: the urge to doom scroll Reddit when you're bored. Instead of opening another tab filled with hot takes and engagement bait, he types three words into Claude—"Teach me something"—and gets a personalized learning session instead.

The Setup

Tunius configured this workflow as a Claude project with custom instructions that prioritize Socratic teaching over direct explanation. The prompt instructs the AI to gauge existing knowledge through dialogue, choose diverse topics from a curated list (Programming, CS, UX/UI/UXR, Cybersecurity, ML, Cooking, Physics, Economics, Psychology, Engineering, and Music theory), and guide discovery through questions rather than information dumps. At each session's end, Claude directs him toward primary sources—websites first, then papers, podcasts, and books—to verify claims and dig deeper.

Why the Socratic Method Hits Different

The critical insight came when Tunius realized plain explanations felt hollow compared to guided inquiry. 'Initially I didn't instruct Claude to use the Socratic method, but that works much better,' he notes. The AI successfully shortcuts basics when it detects prior expertise, adapting difficulty in real-time through conversation. This effectively combines two LLM strengths: non-determinism keeps topics fresh across sessions, while text-based dialogue enables genuine back-and-forth learning rather than passive consumption.

Topics Covered So Far

The diversity has surprised Tunius. Sessions have covered The Allais Paradox (behavioral economics), the physics of consonance (why certain note combinations sound harmonious), and the chemistry of salt in cooking—topics chosen specifically to avoid repetition across previous chats within the project. Claude can list prior conversations to track what's been covered, maintaining variety over time.

Friction Points Worth Knowing

The workflow isn't perfect. Claude often auto-names sessions "Learn something new" based on first user interaction, making history tracking messy. Since no tool call exists for renaming chats, Tunius works around this by asking for session name suggestions at the end and manually renaming himself—a small friction that speaks to how project-level memory still has gaps in current LLM implementations.

Key Takeaways

  • LLMs excel when combining non-determinism (fresh topics) with text-based dialogue (real conversation)
  • The Socratic method outperforms direct explanation because it adapts to your knowledge level automatically
  • End-of-session primary source recommendations guard against hallucination while offering next steps beyond the LLM
  • Project-level chat history enables tracking and diversity across learning sessions

The Bottom Line

This isn't just clever prompt engineering—it's a template for reclaiming idle screen time with intentional learning. Tunius has essentially built a friction-free alternative to algorithmic content that actually makes you smarter, one Socratic question at a time.