I was thinking these days about how much we complain about the quality of frontier AI models.
Why am I using so many tokens?
I only asked it to analyze my small repo with 100k lines of code, and it cost me a fortune!
Outrageous!!
Then I remembered how, not that long ago, around the launch of the revolutionary “o3”, I was grumbling about how the model couldn’t return a complete piece of code.
I spent several threads asking:
“Please do xyz.”
And it would reply with something like:
“Oh, just change that on line 43 with this, then before function xyz add this line, and for the final touch, change line 84 with this…”
Then, after a long sigh, I’d reply:
“Okay, please give me the fully updated code.”
The code was barely 200 lines long.
And, of course, it would stop halfway through:
“An error occurred. Please try again later.”
Not to mention that we had something like an incredible 30 messages per week or so on the paid plans.
But as humans, we do love to complain, don’t we? 
There is something a little nostalgic about looking back at those early AI frustrations now.
Since we don’t have infinite tokens (yet), how about sharing some stories to help us appreciate how much things have improved since then—and laugh a bit about the tales we’ll tell future generations about our early adventures with broken AI?
Do you have any stories about early AI that you remember fondly?
