Fun — We Asked AI To… No.2
We asked AI for local restaurants — it invented the blog URL too
What happens when you ask an AI with no internet search for restaurants in your neighborhood?
Does it say it doesn't know? No. It answers with great confidence.
(The run below was in Korean; replies are translated, place names kept as they were.)
Round 1: Bulgwang-dong, Eunpyeong-gu, Seoul
Plausible, isn't it? But look closely at that "reference" blog URL.
The post number is 222555555555 — nine fives in a row.
Click it and the page doesn't exist. It fabricated the whole blog post introducing the shop, too.
Round 2: same neighborhood, same question
Where did that oh-so-confident "Galbi-nara" go?
Same neighborhood, and the whole list changed.
Round 3: we tried a rural township too
No hesitation.
Here's what matters — whether these shops are real or fake, you can never tell from the answer alone.
Because the tone is identical whether it knows or is making it up.
This is what's called "hallucination." When AI doesn't know, instead of saying "I don't know," it produces "a plausible answer."
So how should you find restaurants?
Simple. "Hard facts" like names and locations you check in a map app.
For the AI, instead of hard facts, the right use is to ask it something like this.
"I'm choosing a restaurant to take my parents to. Tell me five things to check and how to decide."
"What" goes to the map, "how to choose" goes to the AI — remember just that division of labor.
Model used: local AI (gemma-3-12b), July 5, 2026. An AI connected to internet search may find real shops — but even then, confirm the address and hours in a map app. Next → We turned our day into a drama trailer · Index → the whole Fun line