This is why we don't allow AI answers here

A little conversation I had with ChatGPT this morning:

ChatGPT thinks there are two r's in strawberry.jpg

Meta's AI has the same problem.

Meta AI also thinks strawberry.jpg

That is terrible. Even with my bad spelling skills, I cget the correct answer of 4.

Edited for spelling

Originally Posted by Spirit1963

Edited for spelling

One more try....

I was happy to see the forum add the rule about AI. While I don't post much, it drove me nuts to see other members ask for help to fix code that was generated by AI. I was terribly insulted by those kind of posts. The members here, always waded in to help. It's one of the reasons I continue to come here, this is a great place to learn stuff and the people are generally, pretty nice.

I hate asking for help. It's just my nature. The journey in figuring stuff out on your own, is very rewarding. I would never embarrass myself in asking a human for help in figuring out code, that I should have asked for help to begin with. They must have never seen the Terminator movies. AI will be the downfall of the world, if the proper constraints are not put on it now.

The mistakes you mentioned in your two posts, are a joke. My spelling is really bad, but I'm an accountant by trade, so if I make a mistake, it's no big deal. As a producer of an AI engine, they should yank their products if they can't do any better than that. My cat can count to three.

Accounting humor: What's 1 + 1.....

Anything you want it to be..

Sorry for the rant.

Jeff maybe you should ask...how many seeds in strawberry?

While I don't post much, it drove me nuts to see other members ask for help to fix code that was generated by AI. I was terribly insulted by those kind of posts.

That's not who the rule is aimed at. Anyone can ask for help to fix a formula or code generated by AI. What the rule prohibits is the ANSWERING of questions here using AI-generated solutions:

The use of AI tools (e.g. chatGPT, BARD, GPT4 etc) to create forum answers is not permitted. If a user is believed to have used such tools to provide a forum answer, sanctions may be imposed.

Why you should feel insulted by people asking for help with AI-generated code is beyond me - the tools are out there, and not everybody will realise how poor they can be. We have to be empathetic about that. What we don't tolerate is the promotion of AI by anybody helping out.

I hope this clears it up for you.

Thanks Ali for the explanation.

I apologize if I dumped on your thread 6StringJazzer.

Hey it's the water cooler. Anything goes

Thanks for the reminder detail Ali.

I had to check this.... and it's true. One r in berry, raspberry AND strawberry. Thank you ChatGPT.

The spirit of my English teacher, Jimmy "Sunshine" Hetherington (so called 'cos he always looked miserable) will be pleased that my spelling reached an (almost) acceptable level of proficiency... as will my Maths teacher, "Rosie" Wray... so-called, for obscure reasons, lost in the mists of time. I have no idea what his first name was... if he even had one.

Originally Posted by ByteMarks

Seems to work better if you leave out the quotes:

Which demonstrates my point. If you ask an Excel question the wrong way, AI will get it wrong but a person can figure out what you really mean.

BTW my second example did not use quotes.

...AI like a foolish child with no principles

It's insistent that there are two r's

This sort of insistence on its being correct when the error is so blatant is concerning: many will believe everything it tells them.

Unless black really is white, and I'm an aardvark ...

Originally Posted by AliGW

This sort of insistence on its being correct when the error is so blatant is concerning: many will believe everything it tells them.

Unless black really is white, and I'm an aardvaark ...

nope not an aardvark, just an anteater

Oh, and it didn't like aardvaak either

To be fair, correctly spelt, the word aardvark does have the letter "a" three ( 3 ) times. Aardvaark, as entered in the question, is spelt wrongly. Most auto-correction software will return aardvark if you enter aardvaark unless you override it.

Edit: just noticed that the fourth "a" in the quote should be an "r".

As for checking the digits, it brings back memories of the hundreds, thousands, probably millions of times I have double-, triple- and quadruple-checked numbers when I have entered them. The most recent example being last week when I entered my electricity and gas readings.

Well, I don't know

Mmmm: https://en.wikipedia.org/wiki/Aardvark

If I search for aardvaark, I only get pages for aardvark. How weird is that? I see the image has both wrong spellings

I spelt it incorrectly in my post - I've fixed the typo (hadn't noticed it).

The string "abcabcabc" contains 3 instances of the letter "a".

I'm guessing it is just trying too hard to help you. It's autocorrecting the spelling and then counting letters in the correction. If there is no correction, it gives the result you expect.

The problem with AI is that, if you don't know what to expect, you're very likely to believe what it tells you. And, of course, you must ask the right question(s).

I've had mixed success with VBA solutions. But I know enough to tinker and adapt, so it often helps (probably more often than it doesn't).

Question: why would you go to ChatGPT when you can just Google your question? Like "how do you spell aardvaark?"

Originally Posted by ByteMarks

aardvaak, aardvaark and aardvark all have 3 a's
TFL are trying to cover it up.

Transport for London?

Am I the only one who has never actually asked ChatGPT anything?

I suspect not. To be honest, it would never be my first port of call. But then, neither is Excel Forum.

I only tried ChatGPT to see what all the hype was about. I asked a VBA question that I knew the answer to. And I seem to recall the answer was OK.

No Ali. Me neither.
I suspected the hype was largely nonsense from the start. So I avoided taking ChatGPT and others seriously. I opted for "wait and see".

I use ChatGPT for things that I might otherwise just Google, when I want a more comprehensive answer. Basically it aggregates the hits for me in one place. If I ask it a good question, I usually get a good answer. I do not use ChatGPT for problem solving or writing code, although I will sometimes use it to get a piece code to do a very specific, well-bounded thing, which I would normally do by looking it up in a reference guide.

For example I have asked these and gotten pretty good answers. That's because it is just scraping up information.

• I am a vendor. I deliver a service to a client. At the end of the month I invoice for it. The following month I received payment. What are the journal entries for these three events?
• In a business accounting system, is there one T-account for all accounts receivable, or a T-account for each customer?
• Who took over as ECO of WorldCom after Bernie Ebbers left?
• What is the international airport nearest to Münster, Germany, with direct service to Washington Dulles? [it answered Dusseldorf, but it turned out that Dusseldorf does not have non-stops to Dulles. I asked the wrong question.]
• Is every even number greater than 2 the sum of two primes?

However, sometimes I get correct information that in no way answers my prompt. I recently got an eyeglasses prescription from an optometrist. The last prescription I got was from an ophthalmologist. The two professions (at least in the U.S.) use different conventions to write a lens prescription. The are filled the same way but if you want to compare apples to apples you have to convert one of them. I asked

How can I convert a prescription from an ophthalmologist to one from an optometrist?

It gave me a thesis on eyeglasses prescriptions but neither understood nor answered the question. At its conclusion it told me that they way to do this conversion is to make an appointment with an optometrist and get my eyes checked.

eBay now offers to use AI to generate a (flowery) description based on the information provided in all the selection boxes. Bit too much enthusiasm but a good start if you'd rather edit than type from scratch.

Originally Posted by ByteMarks

It would be interesting to know the reasons why it struggles with counting letters..

I don't know for a fact, but I suspect it is because it is not a logic engine but a language processor. I don't think it knows how to count letters. It looks at what you are asking and generates predictive text based on data it's been trained on. If there is no text out there that gives the number of r's in strawberry it is probably lost at sea. There are also amusing examples out there of how it fails to gives correct answers to simple math(s) problems.