Dunning-Kruger vibes from this blog post.

Look into how LLMs work and it's pretty clear why these character counting scenarios often fail if not invoking Thinking/Python Scripting.

undefined

What is your point, exactly? You should state it clearly at the start of the post.

When I asked gemini, it printed the right answer, and it also had a button: "Show code", which I clicked: word = "strawberry" count = word.lower().count('e') print(f"The number of 'e's in '{word}' is {count}.")

Now, the followup (from a comment below): "How many straws in strawberry": there are 0 "straws" (again with the Python code). Similarly, "How many straw in strawberry": 1 straw (again with the Python code, showing it's just trying to do string matching).

Next: Q: "When I saw straw, I mean the object you use to drink liquids through. How many straw in strawberry?"

A: "While the word strawberry contains the letters to spell "straw" exactly one time, there are zero actual drinking objects inside the fruit. Trying to use a berry as a straw would mostly just result in a very messy snack!"

This obvious flaw is immaterial. LLMs are entirely adequate for use cases involving financial gain, like spam and phishing emails, marketing and propaganda.

Follow-up Q: How many straws?

A: One.

5:10 AM here, the writing is a bit messy to be honest and not up to my standards but I wished to upload it anyway because I will have some good comments to read when I wake up. So thanks for reading if you have read it!

And even if you haven't and have just read this comment, that's completely fine too and honestly I don't know why really, but I want to just say that I love you all and I love this community and yes it has its problems but I love you all and I wish you all to have a nice day/night!

Going to go to sleep now. Bye!