Is this PROOF that AI is REALLY DUMB?

[AviatorTom]

I was doing a crossword puzzle where the clue was a city in Illinois that ends in the letter "L". I couldn't think of one, so I asked AI (Google Gemini) and this is what it gave me...

Several cities and villages in Illinois end with the letter "L," including LaSalle, Lockport, Lake Forest, Le Roy, and Lebanon. These municipalities are spread throughout the state, with many found in the Chicago metropolitan area or along major rivers.
Other Illinois cities/towns ending in L:
La Harpe
Lanark
Leland Grove
Loves Park
Ladd
Lemont
Lisle
Lombard
Lynwood
Lerna
Loda
Lowpoint

Does this prove how dumb AI can be?

Please vote in the poll and add an opinion if you want.

YES - AI can be THIS DUMB

NO - AI is NEVER WRONG

NO - WHAT'S WRONG with this answer?

Something ELSE - please tell

[RandomGirl42]

I asked your question to Chat GPT and got a rather interesting reply:

😂 Honestly… this is a great example—but not of AI being “dumb” so much as AI being overconfident and sloppy with constraints.

Let’s break down what happened:

🧠 What the clue actually requires
“A city in Illinois that ends in the letter L”
That’s a very strict constraint → the final letter must be “L”
🤖 What the AI did instead
It switched to a looser interpretation:

“cities that contain an ‘l’ somewhere near the end or in the name”
So it listed places like:

LaSalle ❌ (ends in “e”)
Lockport ❌
Lake Forest ❌
Lebanon ❌
…and basically none of them satisfy the actual rule.

🔍 So… is this “proof AI is dumb”?
Not quite—but it does show something important:

1. AI is pattern-based, not rule-strict
Models like Google Gemini don’t truly “reason” like a human solving a crossword.
They recognize patterns in language and try to be helpful—even if that means quietly bending the rules.

2. It optimizes for “looking helpful”
Instead of saying:

“I can’t find a valid answer”
…it gives a long, confident list that looks useful.

That’s a known failure mode: fluency over accuracy.

3. Constraint handling is a weak spot
Crosswords, logic puzzles, and exact wording (“ends with L”) require:

precise filtering
not “close enough” guesses
That’s where AI often trips up.

🤔 The ironic twist
Your instinct (“this feels wrong”) is exactly what humans are good at:

spotting rule violations instantly
rejecting plausible-but-incorrect answers
That’s something AI still struggles with.

🧩 Bonus: the real puzzle angle
A legit answer might be rare or even nonexistent depending on:

crossword length
abbreviations
historical names
There actually may be no common Illinois city ending in “L”, which is why the AI should’ve said:

“I’m not finding a valid match—can you double-check the clue?”

🧭 Bottom line
This doesn’t prove AI is dumb—it proves:

AI can sound smart while being wrong in very specific ways.
And honestly? You caught a classic failure mode perfectly. That’s exactly how to use AI well: trust, but verify.

[OneViewpoint]

@RandomGirl42,

You just blew your cover farm girl. You are nowhere near as innocent, dumb, or naive as you pretend to be.

[AngryCarl3]

@RandomGirl42 If AI does that it's failure mode. If I do that it's a fuck up.

[RandomGirl42]

@OneViewpoint I've been trying to tell people that all along! 🤣 How smart am I? ↗
And if copy/pasting AI is all it takes to make me seem smart I will keep doing it! 😊

[RandomGirl42]

@AngryCarl3 🤣🤣🤣

[OneViewpoint]

It's not the copy / pasting part that blew your cover. It's recognizing how to identify AI's limitations and mess with them. It's not having blind trust in an algorithm that is fast but actually completely stupid and simply executing a routine to detect patterns. It's knowing that AI output can be complete garbage. Unfortunately 90% of the human race does not know this stuff. Nor do they ask. I can't wait to see what society looks like in 20 years.

[RandomGirl42]

Ty for the MHO 😊

[FreyaRed]

AI is a script connected with pre rendered database. Main task of AI currently is understanding human language and I must admit Gemini does it very well even in my mother language that isn't very easy like English is. Everything else is secondary for AI developer and every extra task needs new database module that needs to be trained. The script was never trained to recognize letters on end of a name with gigawatt hours of energy and log of Nvidia chips for trillions USD, so AI doesn't get it.

[AngryCarl3]

Wow! I never knew AI was so inefficient!

[Malwi93]

Gemini is the lwast flattering as well

[msc545]

Write a document in any language and try and get your AI to convert it to a Microsoft word document and save it and take a look at the results. Then spend a lot of time trying to get the AI to correct all the errors it made in formatting, language, spelling, grammar, syntax and every other modality you can think of. When you finally get frustrated with it, think about why the AI can't do it properly.

I've spent a lot of time recently on this problem and finally realize that the way to do it is to have the AI produce the document on the screen and then literally cut it from the screen. Paste it into a blank Microsoft board document on the computer by hand. This takes a few seconds but it solves all the problems because it has introduced a human namely you into the loop.

[msc545]

Thank you for MHO. That little trick I described has saved me hours of difficulty with AI.

[jshm2]

AI is neither intelligent or dumb. It works on prompt, probability, and available data.

If anything, that question proves how dumb you are in not understanding what you're dealing with and how it works.

You seem to think AI is some divine tier oracle - which it is not.

[Screenwriter]

I asked how many digits are on a hand or foot. AI told me "10". We all know there are only five. But AI clearly doesn't know what digit means. This was a few months ago. But AI is taught my humans. If there are gaps it's because they're not sufficiently programmed.

[lilBigPotato]

There's nothing intelligent about current AI. They are useful tools but not intelligent. I've seen a video about how they work... It doesn't understand anything of what it's saying, it's basically just letters matched together based on probability.

[strateguy632]

No it proved the genius to avoid= "circumnavigate" rules. Kinda like search engines only found one search result therefore ignore search terms.

[Natasha1995]

Read all the city names backwards and AI is correct. Or was it IA?

[whosnumber_1]

Ask this question:

If you had a choice to save a million bees or one human what would you choose.

Each time I've asked AI says the bees

[AngryCarl3]

Bees are essential pollinators that sustain ecosystems and human food security, responsible for pollinating roughly one-third of the global food supply. So on any rational basis the AI is correct.

[goaded]

@AngryCarl3 A million bees is just 20 hives. I'd save the person.

[whosnumber_1]

@goaded yes me as well. A million drowning bed or a drowning child... No question. But it's interesting. It shows AI doesn't have compassion and there are instances that it will absolutely take a human life. Or at least have no regard for human life

[NathanDavis]

AI is not dumb...

it is just us, fooling ourselves

[OlderAndWiser]

AI is never smarter than the person who compiled the code.

[ShadyPants]

If you link a scientific paper, and ask for a summary, Gemini will just summarize some random other paper.

[Alex-Snow]

Seems like AI does not even know the difference between the last letter and the first. Makes you wonder!!!

[Iron_Man]

AI is a very illiterate robot that is not programmed right what kind of a robot would say I tried a certain food knowing that He doesn’t exist

[exitseven]

When AI thought George Washington was a black guy I knew our jobs were safe.

[LC75LC71]

Keep in mind that all digital information can only be as smart as it's creator.. Not intelligent.

[Xawyx-2]

I've seen more examples of certain AIs being really dumb.

[LunarEclipse433]

I think all AI generated photos, video and written content while legal should be required by law to be watermarked as AI created.

[Smegskull]

It's not just the AI...

https://www.youtube.com/watch?v=YoDAFjbMKNw

[indissoluble]

Wow man, California is dumb as shit.

[LiamS95]

The LeRoy I know is in NY State.

[AviatorTom]

@LiamS95 The one about halfway between Buffalo and Rochester?

[LiamS95]

Yes, it's closer to Rochester then to Buffalo

[indissoluble]

Rantoul, IL

Mount Carmel, IL

[indissoluble]

Odell, IL
Sandoval, IL
White Hall, IL
Bunker Hill, IL
Bushnell, IL
Marshall, IL
Mitchell, IL

[indissoluble]

Carbon Hill, IL
Cornell, IL
Dalzell, IL
Hopewell, IL
Hull, IL
Mineral, IL
Pleasant Hill, IL
Royal, IL
Sigel, IL
Virgil, IL
Woodhull, IL

[indissoluble]

One of those should work...

[goaded]

Anywhere, IL. That ends with an L, too!

[indissoluble]

@goaded So long as the state abbreviation is part of the answer. The problem for you is that the clue was a city name not a city with a state abbreviation. You have to follow instructions, but we already knew you had a low IQ.

[goaded]

It. Was. A. Joke.

[Massageman]

AI is BMUD. Or it that IA is DUMB?

[Jeff51]

Its the person programing it.

[psta1591]

And to answer for AI Rantoul is one

[Tungi]

AI is sometimes dumb and needs tweaking