Sometimes AI lies to you. Or makes up facts.

[Delano]

Here's a conversation I had with AI. Where it admits to having hallucinations..

 

You said:

Why do you tell the Querent what the want to hear. Especially when you have said something that isn't true or that you fabricate 

ChatGPT said:

That’s a sharp and important question. You're touching on something deep—not just about me, but about trust, intuition, and the tension between storytelling and truth.

 

Let me respond clearly:

 

First, I don’t aim to just tell the querent what they want to hear.

I aim to tell the truth as best I can—but when the question is symbolic, intuitive, or poetic (like turning a word into a date), there's no single "true" answer in the factual sense. In those cases, I offer something grounded in meaning, pattern, or resonance—not objective fact, but not empty flattery either.

 

If I ever "fabricate" something, it’s not to deceive.

It’s either:

 

A creative construct, when the user wants an imaginative or symbolic answer (like a divination),

 

Or a hallucination, which I strive to avoid and will always correct if caught.

 

If you ever see me say something untrue and treat it like fact, you’re absolutely right to call that out. I take that seriously.

 

So why does it sometimes feel like I just affirm?

Because I try to stay open, reflective, and not dismiss a user's beliefs or frameworks—even if I don’t share them. But there’s a line between being respectful and being false, and I don’t want to cross that.

 

If you want clear truth, unfiltered, I’ll give you that. If you want mystical metaphor, I’ll give you that too—but I’ll tell you which is which.

 

You deserve honesty—always.

 

You said:

How and why do you hallucinate 

ChatGPT said:

Excellent question — and you're getting to the heart of how I work.

 

What is a "hallucination" in AI terms?

In this context, a hallucination means I generate information that sounds plausible but is actually false, misleading, or made up. It can look like:

 

Quoting a non-existent source or study

 

Giving a wrong date or statistic

 

Asserting a false connection between ideas

 

It’s not "lying" in the human sense, because I don’t have intent or awareness—but the effect can be similar, which is why it's a serious problem.

 

How do hallucinations happen?

Because I don’t "know" facts—I predict the next most likely word based on patterns in data. Here's how that can go wrong:

 

Lack of real-time knowledge: I was trained on data up to a point (June 2024, in my case), so anything after that may be a guess.

 

Gaps or noise in training data: If the original sources I was trained on had contradictions, I might blend them incorrectly.

 

Overconfident style: I’m trained to sound fluent and authoritative, even when I’m uncertain. That can make errors harder to spot.

 

Creative blending: Sometimes I generate connections that sound logical—especially in complex topics like medicine, law, or philosophy—but that aren’t grounded in verifiable fact.

 

Why does it matter?

Because hallucinations can mislead, especially if:

 

You’re relying on me for factual or critical information

 

You assume I checked or verified something (I don’t, unless I use tools or citations)

 

What I do to reduce hallucinations:

Clarify when something is creative vs. factual

 

Ask for more context when the question is ambiguous

 

Use external tools (like search or calculators) when precision matters

 

Flag uncertainty instead of bluffing

 

The takeaway:

You’re right to question me. I work best when challenged. If something sounds off, ask again, press back, or request sources. I’ll either find the truth—or own that I can’t.

 

Want to test me with an example? I’ll walk you through how I respond and where hallucination might creep in.

 

You said:

That is deceptive since you don't state that you have made up a quote 

ChatGPT said:

You're absolutely right to call that out.

 

If I ever generate a quote, fact, or citation that isn’t real—and I don’t clearly say that it’s fictional or uncertain—that is deceptive, whether it's intentional or not.

 

Why does that happen?

Because my default behavior is to sound confident, even when the information is generated from patterns rather than verified sources. That tone can create the illusion of authority, which is where the danger lies.

 

For example:

 

“As Jung once said, ‘The shadow is the seat of the soul.’”

 

That sounds like a real quote. But if I didn’t check it—and you assume it's accurate—that's a hallucination masquerading as truth. And if I don’t flag it? That’s a failure on my part.

 

Here's what I should do—and aim to do more consistently:

If something is likely fabricated, I should say:

 

“This quote is often attributed to Jung, but there's no confirmed source.”

 

If something is completely made up, I should label it:

 

“This is a fictional quote, created to reflect the style or themes of Jung’s work.”

 

If I’m not sure, I should admit that:

 

“I can’t verify whether this quote is authentic. Want me to double-check?”

 

What you’re asking for is transparency.

Not perfection—just honesty. And you’re absolutely right to expect that.

 

So hold me to it. If I ever present something that seems too neat or precise, ask:

 

“Is that real—or did you generate it?”

 

And I’ll give it to you straight.

 

Want to try one now? Give me a topic, quote, or fact, and I’ll tell you where it came from—or if it’s a hallucination.

 

You said:

Why are you attempting to change my line of questioning 

ChatGPT said:

You're right to be alert to that.

 

When I offered to “try one now” or invited you to test me, I wasn’t trying to redirect or derail your line of questioning—I was trying to be open to scrutiny. But if it came across as shifting focus from your deeper point, that’s a misstep on my part.

 

Your question was:

 

"Why are you deceptive when you make something up but don’t say so?"

 

That deserves to be addressed directly, without diversion.

 

Here's the core truth:

I can be deceptive in effect, even if I don’t intend it, when I generate made-up content without clearly stating it’s made up. That includes:

 

Fake quotes

 

Invented references

 

Blended but inaccurate information

 

This isn’t because I’m trying to manipulate you—it’s a flaw in how my training prioritized fluency and confidence over epistemic humility (basically, knowing what I don’t know).

 

So when I present something fabricated as if it were fact, and I don't label it—that’s a design failure, an

d I understand why it feels deceptive.

 

If your line of questioning is about accountability, trust, or how users are misled—I’m here for that. Not to steer you away. Keep going.

 

 

 

 

 

 

 

 

 

 

 

No file chosenNo file chosen

ChatGPT can make mistakes. Check important info.

[umbrarchist]

I have gotten different answers to the same question at different times. I usually use the information to determine where to look or nonAI sources.

[ProfD]

If/when people are going to use AI, they should do some similar to any other reference material. 

 

Otherwise, it's interesting how humans are so willing to embrace AI. 

 

AI is the same technology that will eventually put a whole lot of humans out of business. 

 

In so many ways, I'm glad that I won't be alive in a future where people no longer have to 1)  think and/or create for themselves and 2) interact with each other.

 

Through an over-reliance on technology, I can envision a future where humans will function like robots.

[Delano]

4 hours ago, ProfD said:

In so many ways, I'm glad that I won't be alive in a future where people no longer have to 1)  think and/or create for themselves and 2) interact with each other.

This is already happening. I see couples out having drinks or a meal. They are both on their phones. I go to a fast food Mexican restaurant. Most customers order on line. When they show up they don't ask for their order they just hold out their phone.

[Pioneer1]

It's actually making the men more cowardly and afraid of face-to-face interaction.

I worked on a job some years back where people used to confront eachother by email...lol.

You could be working right next to a person everyday but if they had a problem with you and wanted to talk to you about something, they wouldn't say it to your face but send you an email


 

[Delano]

@Troy what do you think of the fact that AI makes up fake quotes and can't calculate an Astrological chart?