When I write something with ChatGPT, I often ask it to critique the ideas rather than the writing. Sometimes I just say, “Criticize this”—more often, I use a more detailed prompt to probe assumptions, gaps, and weak reasoning.
This is useful, but the more interesting step comes next. Once it has finished, I ask it to critique its own critique.
This can be quite revealing. The first response often sounds critical, but it may still be shaped by the way I have framed the piece. It can pick up on my tone, my interests, and the direction of my thinking, and feed something back that fits too comfortably.
The second critique is not necessarily more truthful. It is still the same system responding to another prompt. But it can create a little more distance. It may notice where the first critique was too soft, too abstract, or too accepting of my assumptions.
For me, the value is not that the AI suddenly becomes objective. It does not. The value is that the loop makes the interaction more visible. I can see more clearly how my framing shapes the response, and where the model may be following my lead rather than challenging me.
So the simple sequence is: write, critique, critique the critique.
It is not a method for finding the truth, but it is a useful way of slowing down, testing the thinking, and becoming a little less easily persuaded by a response that sounds plausible.
Knowledge Letter: Issue: 311 (Subscribe)
RSS: Blog Feed
Image Credits: Midjourney
(Public Domain)
