mdh-archive

Recursive gap filling

How to become an expert at anything through recursive knowledge gap filling:

  1. Choose project
  2. Recursively ask model about anything you dont intuitively understand
  3. Explain things back to find gaps in understanding
  4. Recursively ask model about anything you dont intuitively understand
  5. Explain things back to find gaps in understanding

Repeat until you can explain [x] in simple terms. As the great Richard Feynman said: “If you can’t explain it in simple terms, you don’t understand it well enough.”

The power of RGF loops

RKGF loops fundamentally alter the rate of information retrieval and the application of knowledge, effectively compressing years of learning into days.

The process of RKGF is simulating the process of a hypothetical infitently patient tutor that is an expert at every topic that is available 24/7 and knows how to serve information directly at your capability of understanding.

When mistakes are corrected in real-time, the fastest path from 0 to 1 is simply one step at a time (instead of the slow, lethargic path of learning bottom up).

Historically, this top-down recursive gap filling was impossible for schools and universities to offer because it is extremely hard to scale. It would require a dedicated, 1-on-1 teacher for every single student to identify and explain their unique knowledge gaps in real-time.

The only skill you need is to be good at finding gaps in your knowledge, then mapping that to asking a model, but there are ways you can engineer this as well. Here's how you can systematically force your hidden knowledge gaps to the surface:

  1. Start with the Final Product to Expose the Gaps Instead of trying to guess what foundational knowledge you might lack, start by attempting the final, complex task. Ask the AI to write the complete code for a project (like a diffusion model). When you look at the generated results and try to get it working, the specific lines of code, concepts, or bugs that you do not comprehend immediately become your visible knowledge gaps.
  2. The "Wait a Second" Self-Audit You must train yourself to stop glossing over complex information. Peterson notes that the crucial first signal in learning is actively pausing your progress and asking yourself, "Wait a second, do I really understand this part?".
  3. Identify the Absence of the "Click" You can identify a gap by the absence of a specific feeling. Peterson emphasizes that you need to develop a "really strong signal for when it clicks". If you cannot fundamentally explain why something works the way it does, and you haven't experienced that definitive "aha" moment, you still have a hidden gap that needs to be recursively drilled into with the AI.
  4. Use the "AI Feynman Technique" The ultimate way to uncover a blind spot is to try and teach the concept back to the AI. Once you think you understand something, you should tell the model, "this is my understanding of this, is this completely correct?". By explaining your logic back to the AI, the model will clarify your thoughts and explicitly point out "different aspects that you don't understand," perfectly revealing the gaps you didn't know you had.