Data Source: BigQuery public dataset (bigquery-public-data.stackoverflow), Stack Exchange API (api.stackexchange.com/2.3)
Tools: Pandas, BigQuery, Bruin, Streamlit, Altair
Trollercoaster101 on
It is funny how the LLMs still needed stackoverflow to get training and then killed it as a thank you gift.
TOO_MUCH_BRAVERY on
Actually a big problem. Soon troubleshooting knowledge will all be proprietary training data accessible though an LLM subscription.
sssarel on
How can a decline from 200k to 110k questions before the event be called a plateau? At best the plateau should end at 2018, there is a clear downward trend from there, that accelerates further in 2023.
whaaatcrazy on
Curious if this will reduce overall questions to ones that aren’t easily answered making more complicated ones get more visibility
TheGacAttack on
So people couldn’t ever find the Search feature, but they can find Chat GPT.
I guess people just enjoy the act of asking.
Brighter_rocks on
The decline clearly started years before ChatGPT – 2022 just accelerated an already downward trend
Weshtonio on
It looks like it is directly correlated to the Ubisoft stock price.
Faraway-Fire on
It’s not even about asking AI the questions any more. No longer need to. GitHub Copilot (trained on millions of repos) now just corrects / predicts your code directly.
Cut out this middleman need to even ask a question.
Will be interesting to see a version of this graph for Software Engineering stack exchange, where it’s more about the discussions around the edges rather than raw code.
Justryan95 on
For something like a “plateau” it sure looked like a decline.
DManeOne on
SO is one of the web’s most toxic sites for being a user. It is much more effective to speak with an agent than a bunch of passive aggressive neckbeards
HaroerHaktak on
There are examples showing how stack overflow is a toxic environment and asking even a simple question will get you instantly banned lol.
kylelee on
Ironically had to use Stack Overflow yesterday for debugging while using Claude Code.
RedditButAnonymous on
Stack Overflow has always been a necessary evil, its a genuinely terrible site full of the worst kinds of gatekeeping and hostility, of course AI has replaced it, AI doesnt tell you the question is stupid and point you to a similar-but-not-the-same problem that does not help you
snaggyheadshot on
So how do LLM’s solve questions in the future for future new products and or problems? Genuine question. I am guessing they get a lot of information from platforms like this.
namek0 on
Good! Goodbye stack overflow
ForeverYoung_Feb29 on
More like r/dataisterrifying. Stack Overflow is a phenomenal resource of deep knowledge on myriad topics, especially if you find yourself dealing with a legacy codebase. Questions and answers from a decade+ ago are still relevant and useful when you crack open some servlet a long-since retired engineer wrote to fix a strange edge case bug. Hopefully the decline in new questions doesn’t lead to the decline and shuttering of the site because when that goes, so does most of that content. We’re also looking at a possible future gap in expert knowledge where AIs are just answering things based on what’s there from the past and the knowledge base never really grows.
ballrus_walsack on
The llms learned from stack overflow. Where will they scrape their answers from now?
modsaregh3y on
And people are surprised why? SO is a toxic cesspool being gatekept by “seniors” who berated juniors for trying to figure stuff out. Having to read through mountains of threads and oages to try and get a simple answer was backwards.
Sure you maybe learn some nuance going through those threads, but it just isn’t worth it anymore.
A better tool was created and SO didn’t stay with the times. Wish we could berate them for being backwards
malenkydroog on
Makes sense, since ChatGPT won’t respond to a question with “Your question looks like a duplicate….”, and then spend several posts arguing with you about it. 😉
![[OC] Impact of ChatGPT on monthly Stack Overflow questions](https://www.byteseu.com/wp-content/uploads/2026/02/6hd7x0mffulg1-1536x1118.png "[OC] Impact of ChatGPT on monthly Stack Overflow questions")
20 Comments
Data Source: BigQuery public dataset (bigquery-public-data.stackoverflow), Stack Exchange API (api.stackexchange.com/2.3)
Tools: Pandas, BigQuery, Bruin, Streamlit, Altair
It is funny how the LLMs still needed stackoverflow to get training and then killed it as a thank you gift.
Actually a big problem. Soon troubleshooting knowledge will all be proprietary training data accessible though an LLM subscription.
How can a decline from 200k to 110k questions before the event be called a plateau? At best the plateau should end at 2018, there is a clear downward trend from there, that accelerates further in 2023.
Curious if this will reduce overall questions to ones that aren’t easily answered making more complicated ones get more visibility
So people couldn’t ever find the Search feature, but they can find Chat GPT.
I guess people just enjoy the act of asking.
The decline clearly started years before ChatGPT – 2022 just accelerated an already downward trend
It looks like it is directly correlated to the Ubisoft stock price.
It’s not even about asking AI the questions any more. No longer need to. GitHub Copilot (trained on millions of repos) now just corrects / predicts your code directly.
Cut out this middleman need to even ask a question.
Will be interesting to see a version of this graph for Software Engineering stack exchange, where it’s more about the discussions around the edges rather than raw code.
For something like a “plateau” it sure looked like a decline.
SO is one of the web’s most toxic sites for being a user. It is much more effective to speak with an agent than a bunch of passive aggressive neckbeards
There are examples showing how stack overflow is a toxic environment and asking even a simple question will get you instantly banned lol.
Ironically had to use Stack Overflow yesterday for debugging while using Claude Code.
Stack Overflow has always been a necessary evil, its a genuinely terrible site full of the worst kinds of gatekeeping and hostility, of course AI has replaced it, AI doesnt tell you the question is stupid and point you to a similar-but-not-the-same problem that does not help you
So how do LLM’s solve questions in the future for future new products and or problems? Genuine question. I am guessing they get a lot of information from platforms like this.
Good! Goodbye stack overflow
More like r/dataisterrifying. Stack Overflow is a phenomenal resource of deep knowledge on myriad topics, especially if you find yourself dealing with a legacy codebase. Questions and answers from a decade+ ago are still relevant and useful when you crack open some servlet a long-since retired engineer wrote to fix a strange edge case bug. Hopefully the decline in new questions doesn’t lead to the decline and shuttering of the site because when that goes, so does most of that content. We’re also looking at a possible future gap in expert knowledge where AIs are just answering things based on what’s there from the past and the knowledge base never really grows.
The llms learned from stack overflow. Where will they scrape their answers from now?
And people are surprised why? SO is a toxic cesspool being gatekept by “seniors” who berated juniors for trying to figure stuff out. Having to read through mountains of threads and oages to try and get a simple answer was backwards.
Sure you maybe learn some nuance going through those threads, but it just isn’t worth it anymore.
A better tool was created and SO didn’t stay with the times. Wish we could berate them for being backwards
Makes sense, since ChatGPT won’t respond to a question with “Your question looks like a duplicate….”, and then spend several posts arguing with you about it. 😉