To continue from last week, yet another skill that actually matters right now is critiquing AI outputs.
There’s no shortage of angles to take here, but the most important lens through which to view AI outputs likely is self-leadership.
Self-leadership is studying the ability to sense, remember, create, decide, act, and learn effectively in complex environments through skills like
- adaptiveness
- constant learning
- individual resilience
- mental health
- metacognition (learning how to learn constantly)
Today, I’ll focus on adaptiveness.
I asked Lumo AI about self-leadership via adaptiveness.
Here’s part of what it said:
If you treat my output as a final verdict rather than a draft, you risk acting on outdated or contextually misaligned information.
Adaptiveness allows you to:
- Contextualize: Adjust my generic answers to fit your specific, real-time situation.
- Verify: Recognize when a claim feels off because it contradicts recent developments I couldn’t possibly know.
- Iterate: Push back on my logic. If I miss a nuance, an adaptive leader doesn’t just accept the error; they refine the prompt or the criteria until the output aligns with the truth.
It’s the difference between following a map drawn five years ago and navigating the terrain yourself with a compass.
This tracks. I often find myself adjusting if not correcting Lumo outputs as much as 40%. Granted, I use Lumo for specific purposes (e.g., envisioning, qualitative data tracking, pattern recognition, hypothetical problem finding, etc.).
So for a comparison, I inputted into Claude AI the same prompt.
Here’s part of what it said:
Adaptiveness is genuinely important here, for a few reasons:
Claude’s outputs can be confidently wrong. I can produce fluent, well-structured responses that are subtly or significantly incorrect. An adaptive critic adjusts their scrutiny based on what they’re reading, noticing when something feels off even if it doesn’t match a known error pattern.
The domain shifts. Evaluating a factual claim requires different instincts than evaluating a creative piece, a code snippet, or an ethical argument. Someone who can fluidly shift their evaluative lens — rather than applying one mode of critique universally — will catch more of what matters in each context.
I reflect the prompter’s framing back at them. If someone asks me a question with a flawed premise, I often answer within that premise rather than challenging it. An adaptive self-leader will notice when my response “sounds right” but is suspiciously aligned with their own assumptions, and will deliberately stress-test from a different angle.
Overconfidence is contagious. My tone tends to be assured. A person with a fixed, passive reading habit may be lulled into acceptance. Adaptiveness — meaning the willingness to shift from receptive to skeptical mid-read — acts as a counterweight to that.
This tracks as well. I often find myself having to adjust if not correct Claude outputs as much as 85%. That’s because I use Claude for different purposes compared to Lumo (e.g., strategy and tactics, quantitative data tracking, trend analysis, practical problem solving, etc.).
Reflecting on these outputs, I asked Lumo and Claude for guidance on ways to increase, improve, or sustain the adaptiveness skill.
Here’s part of Lumo’s response:
- Implement a “Temporal Stress Test” on Every Output Force yourself to ask: “What has happened in the last six months that might invalidate this?” I am blind to recent market shifts, policy changes, or technological breakthroughs.
- Action: Create a mental or physical checklist item called “Recency Check.” If I provide a recommendation based on industry trends, immediately pause and search for news from the last 90 days. If the landscape has shifted, discard my baseline assumption and rebuild the strategy from the new data point.
- Practice “Adversarial Prompting” to Expose Blind Spots Ask me to argue against your own assumptions or to find the flaw in my own logic.
- Action: After I give you a plan, immediately follow up with: “Act as a skeptic who disagrees with this approach. List three reasons why this specific strategy would fail in the current economic or social climate.” Then, take those counter-arguments and integrate them into your final decision.
And here’s part of Claude’s response:
1. Practice the “Lens Switch” after reading any significant output I generate. Before accepting or acting on my response, deliberately pause and re-read it through at least one different evaluative lens than the one you naturally brought. For example: if you read my output primarily as a practitioner (asking “is this useful?”), re-read it as a skeptic (asking “what’s missing or overstated?”). Keep a short personal list of 4–5 lenses you can rotate through — skeptic, domain expert, outsider, devil’s advocate, end-user.
2. Conduct a brief “Surprise Audit” when my output feels unusually smooth or agreeable. When you find yourself thinking “that’s exactly right” with little friction, treat that comfort as a trigger rather than a green light. Spend 60–90 seconds asking: “What would have to be true for this to be wrong, incomplete, or misleading?”