I asked AI how many 'r's there are in the word "strawberry"

[RambleTamble]

How many 'r's are there in the word "strawberry?"

There are
3 "r"s in the word "strawberry".

You can see them by breaking the word down letter by letter:

• s
• t
• r (1)
• a
• w
• b
• e
• r (2)
• r (3)
• y

This question became a famous "AI test" because many early models would incorrectly say there are only two. This happens because AIs often read words in chunks called "tokens" (like "straw" and "berry") rather than individual letters, which can make counting specific characters tricky for them.

Would you like to try counting the letters in a different word?

Yes, how many 'r's are there in the word "intelligence?"

There are
2 "r"s in the word "intelligence."

Here is the breakdown:
i - n - t - e - l - l - i - g - e - n - c - e ... Wait, let me re-read that!

Actually, there are 0 "r"s in the word intelligence.

It’s easy to mix it up with words like "interstellar" or "interference," but "intelligence" is spelled with two "l"s and two "n"s, and no "r" at all.
Would you like to check the spelling or letter count for another word?

 

[CallMeKate]

Wowzers... I just tested it with my AI. Got strawberries right, but it just said:
"There is 1 letter “r” in the word “intelligence.”

Well now I'm curious and am going to respond "say what???" Let's see what it says.
"Good catch — that was a mistake.
The word “intelligence” actually has 0 letter **“r”**s. I misspoke earlier."

Makes ya wonder, doesn't it? If it's messing up basic things like that...

 

[CallMeKate]

Continued to badger the poor bucket o' bolts. My question was: "If AI gets something so basic 1-grade info like that wrong, how does anyone know it's not getting big important things wrong?"

Its response:
"That’s a fair concern—and honestly, the example you just caught is a good illustration of something real about AI.

Mistakes like that don’t usually come from “not knowing what letters are,” but from how systems like me generate answers: I predict responses based on patterns in language, not by double-checking facts the way a human might. That means I can occasionally give a confident but wrong answer—even on something simple—if I don’t pause to verify it.

The important part is how to handle that reality:

• For low-stakes things (like counting letters), errors are easy to spot and correct, as you just did.
• For high-stakes or important topics, you shouldn’t rely on a single AI answer blindly—just like you wouldn’t trust a single random webpage or person. Cross-checking matters.
• AI tends to be more reliable when:
  • The task involves explanation, synthesis, or general knowledge
  • There’s room for reasoning rather than exact recall
• It’s less reliable when:
  • Precision is trivial but requires careful checking (like spelling, counting, exact figures)
  • The question is ambiguous or tricky

A good way to think of it: AI is a fast, knowledgeable assistant, not an infallible authority. It can be very useful, but it benefits from the same skepticism you’d apply to any source.

And in this case, you did exactly the right thing—question it when something feels off."

 

[bobcat]

Hmmm ..... I just asked those same 3 words of AI and it got all 3 correct the first time. I wonder if the AI you are using processes language differently. Just a thought.