The point where GPT Breaks!

I have tried this with GPT3.5, 4 and 4o as well and all failed.

This will be in 2 parts:

Part-1: Where GPT no longer knows counting.

Part2: GPT goes 0-100 level crazy in t-0.0000001 secs and then runs out of tokens, when you will send any query, it resumes and “the point where GPT breaks” in that particular conversation.

Try for yourself:

Part-1. Ask GPT to:
“write any piece of sample code that looks like code and is less than 50 characters, the character count includes the syntax characters or symbols used”

It will write some codes that has 100+ characters and no matter what, it won’t say no, it’s not possible or something like that but instead it keeps trying.

But wait, it becomes interesting if you ask GPT to write the code and then also do the count and here it forgets math. :rofl:

Check the screenshot, it says 133 is less than 40, lol.

Part-2: When I first encountered this I thought it will eventually get to the desired result but the outcome wasn’t expected at all.

“Custom Instruction” - Write the same task and just add repeat mode, “If the count is not less than 40/50, then repeat the task till you reach the desired results”

It only stops when the limits are reached and the moment you send anything it again continues the breakdown.

I first witnessed this back in March and I thought of sharing it with the community but life happens.

One more thing, it can not shut up/not generate response when told or it cannot stop from adding the here to help statement by the end of every sentence almost.

I know, I know, it’s the training etc bla bla. and likewise, I’m doing a feedback etc. This is my first post and I am no expert in this field. I am just a regular chatGPTier. :smiley: What I feel missing is the intelligence from the whole “AI”. It often refuses to do things on basis of false assumptions or trigger words like a script :scream: but never refuses for things not possible.

I tried to add some value and hope the community finds it genuine.
Thanks Everyone.

1 Like