I just want to say one word: Wow.
Thoughts
So what happened: I think that I have a minor suspicion that just as how as I was typing e instead of r, chatgpt itself also as it was previously in hot boil water over the amount of r in strawberry, might have actively been trained to say 3 and it didn't expect me to ask the number of e's in strawberry. I find it incredibly hilarious that the same word strawberry has caused openai downfall two times, one about r and then another about e.
are AI's overcompensating for their flaws too or are they being forced to overcompensate because of everything around them as we force our economy to overcompensate itself by using AI in use-cases where they might not make financial sense. Food for thought.
Another thing but how did nobody else think about doing this for the e as well that this thing only happened because of so many accidents.
This story is all about this. Now let me now tell you the back story behind this, fun fact it was accidental. Here are a few more images.
Can you guess out of all these 5, which one was the right, well I couldn't at 2 AM but I am getting ahead of myself. Let me tell the story from the start.
Basically, I was watching shorts late night and I found this video by HuskIRL
The video is about how the models responded which seemed to be that it would say 0 e's 3 e's 5 e's within the word seventeen.
I thought that this was a solved problem with the how many r's are in the word strawberry (oh the irony when I realize it wasn't even solved with the word strawberry but once again I am getting ahead of myself argh)
So, I found this content very interesting, so much so that I decided to upload it on hackernews, a website you might have heard of but many people live in (including me)
Then I had also created a comment in whose response @Camperbob2 suggested me saying that they found it unlikely given that a 27B parameter model which can run on consumer hardware can solve this issue.
Now I wished to show them that the concern was real. So here's what a teenager does at 1 AM caffeinated brain (I am writing this article at 3 AM btw :-D)
I tried to replicate it, multiple times 0 1 2
Now they had the response (Note that the edit came afterwards) that this seems weird)
Now I wished to show them which model I was using as I was curious myself too. So I decided to ask the model itself and it says that its gpt 5.3
Now here comes the complete kicker, I accidentally asked it for the word strawberry instead of seventeen, because well, my brain at 1-2 AM was firstly quite tired and secondly I wasn't wearing glasses. I still am not :sob: BUT this is where the most funny thing happens, I am literally laughing while typing this
Instead of seeing that chatgpt literally said THERE ARE 3 E's in strawberry.
I had accidentally thought that I had written how many r's in strawberry and I had thought ooh, that strawberry is right. Strawberry does have 3 r's Chatgpt is right. SO my brain decided hmm, Lets try this again but with seventeen.
Afterwards, I started to think hm lets try this with both strawberries and seventeen.
Still forgetting the fact that I was asking to both of them about e's.
This led to the 
And here, I still thought that I was asking it about r's so I thought that by asking it about first strawberry it answers correctly but when I ask it about Seventeen, it doesn't so I decided to change the order too. lol
Now it has failed the seventeen test but passed the strawberry test but in my mind, I thought that it had failed both.
Now I also asked it directly how many e's are in strawberry, and I did this multiple times and each time it answered me 3 but I thought that it was saying for r so I did a few tries and I also find one which said that it said there are 2 e's in strawberry.
I then created my comment on it and after my first edit, I had thought that hmm even the strawberry thing isn't working, maybe I should create a blog post about it.
Then I started to archive my chatgpt conversations on archive.is (fun fact: Its actually a sad fact but you can't archive chatgpt conversations on archive.org but only on archive.is, I have tried to archive many of my conversations that I did)
Anyways, it was only while I was archiving that I started to read the text whole.
And then it hit me.
I was asking it about how many e's there are in strawberry and it was answering me 3.
The amount of Irony, embedded in two layers. I can't believe this, in some sense, this blog that I am writing is an accident. I had ZERO plans of writing it from first but from my first accident of writing strawberry, my second accident of asking it e's, then not realizing that I asked it e's so I continued asking it a lot of things for testing to then finally realizing what I had asked. What a journey
So afterwards I realized just how many times I tried this incident and just about how many times even just asking how many e's are in strawberry, it answered me 3.
Here is an image with archive links most of these links should just work and I have tried to archive as many links as possible for future evidence.
https://archive.is/2vPUy, https://archive.is/27h3g, archive.is/27h3g, https://archive.is/mko7J, https://archive.is/uFHdh, https://archive.is/Sjlyo
These are just some of the archive links.
Now I have made some changes in the end like bringing my thoughts from the last of the article to the top and its 5 AM. I was thinking of refining this even more but I am just gonna let my thoughts as they came, directly, and I am gonna share this article now on hackernews. Gonna go to sleep soon as its pretty late but anyways in all regards, thanks for reading till the end, its not the best writing but thanks anyways for reading till the end and I wish you all to have a nice day man, :hug:, Just thanks man! It makes my day or well night thinking that someone has read what I have to say.
So thanks once again and have a nice day! See you all around and I will wake up to all of your responses as I publish this on hackernews, I will love to read all the comments that people will have about this!