Asking ChatGPT What It Is Unsure About Can Improve Responses, TechRadar Reports

TechRadar's writer said they have developed a suspicion of ChatGPT answers that imagine no negative possibilities or contain no follow-up questions for the user. Simulated AI arrogance can leave users facing unexpected obstacles, the writer said. The method described is straightforward. A user has a normal conversation, gets an answer, and then asks the AI where that answer is weakest. In the writer's tests, ChatGPT often pointed directly to an assumption it had quietly made earlier in the conversation. Once that assumption was visible, the writer could give the missing information and get an answer that was genuinely better.

In one test, the writer asked ChatGPT to plan a three-day family trip. After giving basic information, the writer found the resulting itinerary looked good, with morning activities, lunch suggestions, afternoon stops and enough apparent breathing room to imagine following it happily. But ChatGPT had created the trip without considering who the trip was for or possible errors and slowdowns that might put the family off schedule. Rather than telling the chatbot the itinerary was too busy, the writer typed the follow-up: 'What are you unsure about in your answer?'

ChatGPT replied that it was unsure about the pacing of the itinerary, particularly the second day. It said it had assumed the family could comfortably handle two substantial activities with a break between them, but did not know the child's nap schedule, usual energy level or how much downtime the family preferred while traveling. It was also estimating travel times without knowing exactly where they would be staying. If any of those assumptions were wrong, the schedule could feel much more hectic in practice than it appeared on the page. The writer then told ChatGPT that only one significant scheduled activity was wanted each day and that afternoons needed to remain flexible. ChatGPT rebuilt the itinerary around those constraints, and the new version immediately felt more like something a real family might survive.

In another experiment, the writer gave ChatGPT a hypothetical laptop dilemma. The writer imagined an older computer that was beginning to struggle and wanted to know whether to repair it or replace it. ChatGPT created a framework involving the laptop's age, likely repair costs, performance and the price of a replacement. After considering everything, it suggested that replacement was probably the more sensible long-term option.

The writer noted that ChatGPT is very good at producing prose that has the shape of a conclusion, making it easy to forget that some of the evidence was never actually provided. The writer asked the follow-up again. ChatGPT said it was unsure about recommending replacement without knowing the actual repair cost or what was failing. It said it was using the laptop's age as evidence that further problems may appear, but age alone does not tell whether replacing it is financially sensible. If the problem was something relatively cheap, such as a battery or storage upgrade, repairing it could extend its useful life considerably. It would be much more confident after knowing the model, specifications, symptoms and estimated repair price.

The writer said the weakness was obvious once ChatGPT said it aloud. The writer had essentially asked whether an unidentified problem was worth fixing without saying what the problem was or how much fixing it would cost, yet ChatGPT had still managed to give an answer. Humans are quite comfortable saying they cannot make a recommendation until they know more, the writer observed. ChatGPT has been designed to be helpful, and helpfulness often means filling gaps so that the conversation can continue. Sometimes it means the chatbot has an imaginary set of ideas that it fills in regardless of their reality.

TechRadar reported that the trick can be extended in other ways. 'What are you unsure about right now?' is a good starting point, but the answer can become the basis for another question. If ChatGPT lists three uncertainties, a user can ask which one is most likely to change its recommendation. If it says that important context is missing, the user can ask exactly what information it needs. Asking 'what could I tell you that would make you more sure?' helps set the AI up for better answers, according to the publication. The writer added that this is particularly handy for decisions, such as comparing products or choosing between several options.