Dear Stack Overflow denizens, thanks for helping train OpenAI's billion-dollar LLMs Stack Overflow, a community-driven Q&A site, and OpenAI, maker of AI models, have agreed to work to improve each other's products, the latest deal in a series of tie-ups to feed machine learning models' thirst for data. The two organizations characterized the partnership on Monday as a way to "strengthen the world's most …
" was planning to retire to a punnery but it would be too conventional."
When Hamlet said to Orphelia get thee to a nunnery I believe in Shakespeare's time his audience understood a nunnery to be the premises which in Terry Pratchett's words, offerred negotiable affection (at reasonable rates.)
So in today's world might not a punnery be less conventional than a berth in the Fool's Guild (Guild of Fools and Joculators and College of Clowns- founder J-P. Pune) and slightly more engaging?
For years, vonC has been one the most productive and highest upvoted contributors on SE, posting marvelous, detailed, helpful answers with millions of accumulated up votes. If AI is an intelligent human equivalent trained on SE, vonC is its foremost teacher, to whom it knows it owes oh so much. When AI trained on SE speaks, the ghost of vonC speaks. Moreover, SE/Google/OpenAI all know that.
So what happened is that they made a deal with vonC, offering him eternal life in exchange for shedding his human body and self-uploading his spirit into ChatGPT4.0. The King is dead. Long live the King.
And in exchange, Stack Overflow can use OpenAI to generate answers?
Seems like this dims the incentive for further participation by humans, and thus OpenAI will have less to scrape in the future. I assume its answers will be identified in some way (even if we can't see it) so that it won't eat its own shit and cause model collapse.
They need to train it and SO has lots of metrics to value if an answer is good or bad.
But they have a major problem with old outdated answers with massive score that always pop up first. I think both parties will benefit. SO can probably get a much better match and some much needed money infusion.
"But they have a major problem with old outdated answers with massive score that always pop up first."
Shirley they'd take the age of the answer into consideration too? In the world of "coding", the age of the answer is often a seriously important metric.
Never had much joy whenever a query misled me to Stack Overflow.
Signal to noise extremely low for most stuff I was interested in and a fair bit of pointless argy-bargy.
The name is bit of a give-away - a stack overflow invariably corrupts data in a previous activation record and/or overwrites the return address and then you are off to the boondocks.
Some of the code I have looked at was just plain wrong and given AI's penchant for hallucination I dread to think of the future perils facing the cut and paste coder. Hint: Get some decent texts on algorithms - discrete math. for computer scientists is pretty useful as is a bit of formal logic and some exposure to formal methods in software development.
Software company whose major cost is programers seeks to replace programmers with machine that works for electrons and has constant, near zero overhead when not in use.
Also, surprised that company that makes a living helping programmers is having issues... I am sure it is unrelated.
As a meatsack, I am so glad I'm on the, safe for now, hardware and arch side of things. But I hear chatGPT is learning layout and xkt design.
I was tinkering with a LLM last night, asked it to write a function to find the nth digit of pi. It wrote a function that converted the built in pi constant to a string and returned the nth character. It then proceeded to have a stack overflow comments section argument with itself about how wrong the solution was and proceeded to start arguing about the question. It was most entertaining. It was clearly trained on stack overflow. I believe it was a codellama derivative.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
