I used OpenRouter’s playground tool to see what OpenAI’s new decision model – GPT-6 Luna Decisions looked like. I gave it a question and a true/false format. Above you will see the result. I found this to be fascinating. From my experience, there is no model that can detect fallacies in a manner I would deem remotely capable. Fallacies are slippery. Many times they are constructed to sound reasonable or right, but under close inspection are not. They are a fine way to create a Trojan horse that smuggles lies into discourse. Sometimes, they are so subjective that any reliability claim is false simply on the basis of what you might call a matter of opinion. So, no human can reliably detect fallacies. No machine can reliably detect them, either. It’s just not something that can even be tagged possible in the truest sense. So, any model that thinks there is a strong potential for any model whether itself or other is arguably dangerous. This hubris is both humorous and frightening. I’m not even sure how to feel about it. But, after having been immersed in this question for long enough to learn that there is no flawless oracle in the world as we know it, I can say with confidence that the statement is fact, not fallacy.
