As a longtime editor, I treat ChatGPT the way I would a talented reporter bringing me an apparently brilliantly researched story.
If a claim really matters, I ask what evidence or sources support it, and then I check them myself. I’ve also asked it to give me a confidence estimate whenever it’s stating facts, and it’s become quite good at distinguishing between what it knows with high confidence and what is more uncertain.
I push back whenever something sounds surprising, implausible, or just a little off. One of the strengths of AI is that it usually responds well to skepticism—it will often qualify, revise, or even retract a claim when challenged. And I ask lots of follow-up questions. The conversation is where the value comes from, not simply the first answer.
Over time, I’ve reality-tested its advice across a wide range of practical subjects—from managing my badly deteriorated but still remarkably serviceable knees to keeping my swimming pool in better shape than I ever thought possible. What I’ve found is that the back-and-forth process of questioning, testing, and refining has been consistently useful.
One more habit I strongly recommend: have another AI model critique an important answer. Different models have different strengths and blind spots, and asking one to look for errors or weak reasoning in another’s response is an effective form of quality control.
Love this info, and you’re right to wonder how you’d have ever come across it in the past.
The NY Times had an article today about AI's incomplete reliability: “We’re Only Starting to Grasp the Pitfalls of Using A.I. at Work” Scholars say the “unknown unknowns” of using artificial intelligence in the workplace may be undermining the technology’s advertised benefits. https://www.nytimes.com/2026/06/29/business/artificial-intelligence-workplace-consequences.html
What steps do you suggest to verify AI's accuracy?
BTW - wishing the best for all of the birds.
As a longtime editor, I treat ChatGPT the way I would a talented reporter bringing me an apparently brilliantly researched story.
If a claim really matters, I ask what evidence or sources support it, and then I check them myself. I’ve also asked it to give me a confidence estimate whenever it’s stating facts, and it’s become quite good at distinguishing between what it knows with high confidence and what is more uncertain.
I push back whenever something sounds surprising, implausible, or just a little off. One of the strengths of AI is that it usually responds well to skepticism—it will often qualify, revise, or even retract a claim when challenged. And I ask lots of follow-up questions. The conversation is where the value comes from, not simply the first answer.
Over time, I’ve reality-tested its advice across a wide range of practical subjects—from managing my badly deteriorated but still remarkably serviceable knees to keeping my swimming pool in better shape than I ever thought possible. What I’ve found is that the back-and-forth process of questioning, testing, and refining has been consistently useful.
One more habit I strongly recommend: have another AI model critique an important answer. Different models have different strengths and blind spots, and asking one to look for errors or weak reasoning in another’s response is an effective form of quality control.
Very interesting. Please keep us posted on the progression of "your" hawk toward maturity.
Just posted an update
Incredible.
OK so now you know we're going to need weekly updates on Junior Hawk, right?
Let’s just hope they don’t involve the mangy fox who also frequents my backyard.