I taught Chat GPT to distrust itself, and suddenly it stopped hallucinating | Tech Radar

Overview

News, deals, reviews, guides and more on the newest computing gadgets

Start exploring exclusive deals, expert advice and more

Details

Unlock and manage exclusive Techradar member rewards.

Unlock instant access to exclusive member features.

Get full access to premium articles, exclusive features and a growing list of member rewards.

I taught Chat GPT to distrust itself, and suddenly it stopped hallucinating

Making AI skeptical of AI answers helps keep it honest

When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.

Anyone who uses Chat GPT or other AI chatbots eventually encounters the confident hallucination. The AI will explain a nonexistent feature, invent a quote, or describe a restaurant that closed during the first Clinton administration.

That's because large language models are designed to produce plausible-sounding responses quickly. That ability is what makes them useful, but it also creates the perfect conditions for hallucinations. The chatbot wants to keep the conversation moving smoothly, so it often fills in gaps with fiction if it's convenient.

I have recently started adding an addition to any of my prompts that ask for facts. I essentially make Chat GPT as skeptical of its answers as I often am. I append this to the prompt: “Act as a hostile AI auditor and assume unsupported specifics are false by default. Mark all uncertain, inferred, or weakly supported claims clearly.”

Using Chat GPT for Iran war news changed how I trust information

I stopped asking AI for answers and started asking for frameworks instead

The wording sounds dramatic, but being so emphatic has proven the best way to ensure Chat GPT follows through. With the additional lines, Chat GPT suddenly becomes more cautious, more analytical, and far more willing to admit uncertainty.

The hostile auditor lines change Chat GPT's tone to one of eagerness to prove its reliability. I tested it while planning a weekend trip. With the standard prompt, Chat GPT had its usual breezy confidence and produced itineraries that I would say were 80% useful and real.

When forced to audit itself, I saw a lot more caution, with sentences like: “Several train schedule details may be outdated or inferred from older timetable patterns and should be verified directly with the transit provider.”

It also flagged one restaurant recommendation with the warning, “Current operating hours and reservation availability could not be independently confirmed.”

The response felt dramatically more trustworthy because of those caveats. The same thing happened when I used the prompt for a theoretical need to fix a noisy dishwasher that is making an unpleasant grinding sound during its wash cycle. Under normal circumstances, I would get a single conclusion and insistence that I start with the assumption of one thing as the problem.

With the hostile auditor instruction added, the tone shifted. Chat GPT wrote: “A failed pump is one possible explanation, but the symptom could also result from trapped debris near the impeller or loose spray arm components. Additional inspection would be needed before assuming component failure.”

Even simple household questions become easier to evaluate with the prompt in place. I asked Chat GPT whether an air purifier would be large enough for my office.

Instead of immediately declaring that it was ideal, the chatbot responded, “Coverage estimates vary depending on ceiling height, filter condition, and real-world airflow.” That cautious wording prevented me from treating a marketing claim like a laboratory measurement.

The prompt does not magically eliminate hallucinations completely, though. Chat GPT can still misunderstand context, rely on outdated information, or misinterpret vague instructions. But it becomes far more transparent about weak spots in its reasoning. Teaching AI to distrust itself may end up being exactly what makes it more trusted.

Follow Tech Radar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.

➡️ Read our full guide to the best business laptops

Best overall: Dell 14 Premium
Best on a budget: Acer Aspire 5
Best Mac Book: Apple Mac Book Pro 14-inch (M4)

Eric Hal Schwartz is a freelance writer for Tech Radar with more than 15 years of experience covering the intersection of the world and technology. For the last five years, he served as head writer for Voicebot.ai and was on the leading edge of reporting on generative AI and large language models. He's since become an expert on the products of generative AI models, such as Open AI’s Chat GPT, Anthropic’s Claude, Google Gemini, and every other synthetic media tool. His experience runs the gamut of media, including print, digital, broadcast, and live events. Now, he's continuing to tell the stories people want and need to hear about the rapidly evolving AI space and its impact on their lives. Eric is based in New York City.

You must confirm your public display name before commenting

1 Massive Lithium deposit potential worth $1.5 trillion found in Oregon — a huge cache could massively strengthen the US stance in building items such as smartphones, but environmentalists urge caution in acting too fast

2“I think it can be a point of no return” — experts issue warning as Chinese EV manufacturers hunt for European manufacturing plants

3I think therefore I am: cutting-edge AI used to 'revive spirit' of French Mark Twain and spawn a new sharp satire more than 350 years after his death

4'This work is a glimpse of what is coming': Security team lays out how Anthropic Mythos helped build a working mac OS exploit in five days

5 It's time to ditch your takeout coffee habit — I'm a trained barista, and these are the top 3 coffee makers I recommend for cafe-quality lattes at home

Tech Radar is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site.

Key Takeaways

News, deals, reviews, guides and more on the newest computing gadgets
Start exploring exclusive deals, expert advice and more
Unlock and manage exclusive Techradar member rewards
Unlock instant access to exclusive member features
Get full access to premium articles, exclusive features and a growing list of member rewards