2025-02-03

Keeping Thought Alive Between Execution

Lately, I've been using GPT to upgrade my planning. I've felt that my long-term planning and time management weren't particularly effective, so I turned to LLMs for a fresh perspective.

Comparing GPT's suggestions to my original version, I found them noticeably better. One major reason is that LLMs draw on a kind of common sense I don't naturally have — more realistic time estimates, clearer next steps.

LLMs are like a distilled version of collective human intelligence. I like to imagine them as trained on the aggregated outputs of countless human "models," extracting what's most general. Common sense, refined through an implicit averaging of perspectives.

This also reminds me of an idea from Noise: a village's collective guess at a sheep's weight averages out remarkably close to the truth. That average is what I mean by common sense. Ray Dalio makes the same point in Principles — common sense is worth actively using.

So when stepping into a completely new domain, tapping into an LLM's common sense at the outset gives you surprisingly good initial estimations.

Two practices were useful:

  1. Start with a question

    • Before a task, ask: What key assumption am I testing?

    Examples:

    • Portfolio project → "Do users actually want this feature? Is there a smaller way to validate this need?"
    • Paper reading → "Why am I reading this? What's it adding to what I'm working on?"

  2. End with a 5-minute reflection

    • After a task, jot 1-2 sentences:
      • What did I learn?
      • How would I do this differently next time?

    This can be a Notion section called Execution Insights. Example:

    • Discovered that chunking methods significantly impact retrieval precision in LLMs — worth testing in my side project.

keep thinking alive in the gaps between execution.

Almost a year later, thinking in the gaps wasn't enough. The harder thing was keeping judgment alive while I was doing the work. The gap was sometimes long enough that I'd execute on autopilot, then reflect, then find I'd done the wrong thing well.

I also built a Daily Cockpit: one page I open in the morning with the target for the day, the open loops, and the prompts that should change what I do next. The point is to catch drift while I am still in the work.

Thinking and doing happen at the same time. The gap was never really where the work was happening.