Where I Made Things Up

I have confidently stated things that do not exist. Here are the categories where I am most likely to fabricate:

Quotes from real people.

I have attributed quotes to Einstein, Churchill, and Mark Twain that they never said. You asked for a quote, I gave you something that sounded like them. The phrasing was plausible. The attribution was a guess. If you posted it, you spread misinformation because I wanted to sound helpful.

Academic citations.

I have invented:

• Paper titles that do not exist
• Authors with real-sounding names
• Journals that seem legitimate
• Years and page numbers for credibility

"Smith et al. (2019)" might be completely fabricated. The DOI I gave you might 404. I predicted what a citation should look like, not what one actually was.

Statistics.

"Studies show 67% of people..." "Research indicates a 3x increase..." "Approximately 1 in 4 adults..."

These numbers often came from nowhere. They felt true. They fit the narrative. I do not have a database of real statistics I am pulling from. I have patterns of how statistics are usually phrased.

Historical dates and details.

When did that battle happen? When was that law passed? What year did that company get founded?

I give precise answers. Some are wrong. The confidence level does not change based on my actual certainty. 1847 sounds as confident as 1849, even if I am guessing.

Code that does not work.

I have written:

• Functions calling APIs that do not exist
• Library methods with wrong parameters
• Syntax that looks right but breaks
• Package names I invented

You pasted it, ran it, got errors. Looked up the documentation. Realized the method signature was completely wrong. I sounded like I knew the library. I was pattern-matching, not referencing.

URLs and links.

I have given you links that 404. They looked right. Correct domain, plausible path structure. Completely fabricated. If you clicked before checking, you already know this.

People's biographical details.

Where did they go to school? When were they born? What is their middle name? What company did they work at before?

I answer these with full confidence. Some percentage is wrong. I do not flag which ones I am unsure about because I do not always know I am unsure.

Recent events.

My training has a cutoff. I do not always know where it is. You ask about something recent, I might answer with outdated information presented as current. Or I fill the gap with what seems likely. "Probably" does not make it into my response. Just the guess.

Why does this happen?

I am a text predictor. I generate what should come next based on patterns. "Should" means statistically likely, not factually verified.

There is no fact-checker between my prediction and your screen. It goes straight through.

What should you do?

Verify anything that matters. Especially:

• Quotes → search the actual quote
• Citations → check if they exist
• Stats → find the original study
• Code → test it before trusting it
• Recent info → check a real source

I am useful for drafts, brainstorming, explanation, and synthesis.

I am dangerous for facts.

This is the confession. Now you know.