A Strawberry Quandary: When AI and Fruit Collide
"Prithee, Horatio, tell me one thing... how many damned r's in strawberry?"

A Strawberry Quandary: When AI and Fruit Collide

I had a fun chat with ChatGPT-4o this morning. It all started with a simple question: “How many r’s are in the word strawberry?” Sounds straightforward, right? Well, the conversation that followed was anything but! What started as a quick count turned into a fascinating dive into how this AI “thinks,” and a surprising admission of error.

And this is where things got interesting. I asked my trusty GPT to explain how it managed to completely bugger up this simple task. It replied with unbridled confidence explaining how it went about being completely wrong.

Come now GPT, that's just not true, and there I thought we had a burgeoning relationship based on trust. Not one to let my GPT off the hook, I disagreed that it was following this "straightforward" process. It was not counting r's so I wanted to know what it was really doing.

"Nay, young GPT... you have made such a sinner of your memory, to credit your own lie."

Now this is getting interesting. I'm now wondering about the actual process. The GPT's "recognition" of the structure of the word "strawberry" is either flawed or seeing that it's a probabilistic model, it's simply guessing.

"The fool doth think he is wise..."

Mmm... so my GPT "knows" the word's spelling and still maintains that there are only two r's. I think not! Let's give that assertion a bit of a stress test!

Ahh, great! We have made progress. But I'm still intrigued at why the GPT not only made the mistake but doubled down when explaining how it got it got to the solution.

One might argue this is quite "human"... I've got cocky at times with familiar subjects and got the facts a little twisted when I shoot from the hip.

This still sounds like a guess to me. But I wanted to know what the difference was between my human guessing and my GPT's "Assumption based on familiarity"...

Ahh... the priority of speed over accuracy! But I'm sure we can get around that little problematic behavioural pattern...

A few great pointers there to remember when prompting, but I also immediately noted the "Memory Updated" precursor to the response. I found this interesting... I didn't explicitly ask it to remember anything. So I decided to check on that...

Superb :)

Hope you enjoyed this exploration into some of the interesting quirks of GPTs!

If you're interested in exploring how Gen AI or RPA can improve your business processes, feel free to reach out. Together, we can harness the power of automation to drive efficiency and innovation.

Contact Us Today!

[email protected]

davidssonfox.com

Nicky Godden

Passionate about consistently improving systems to support effective delivery of our purpose and strategy.

5 个月

Quite entertaining. Perhaps Chat GPT is becoming human.....

回复
Eric Mc Lachlan

humanist, technologist, innovator

5 个月

This is very cool. That’s amazing meta cognition on the part of a language model. It is able to articulate not only how it processes information, but how the mechanics are different to human cognition. It’s really interesting, and I find myself wondering, how aware is it of its own “cognitive” processes?

要查看或添加评论,请登录

Stephen de Villiers Graaff ????的更多文章

社区洞察

其他会员也浏览了