In the tradition of posting meta posts for feedback, in the hopes that it'll be conveyed to the relevant people, and at the very least, have a space to discuss the most recent blog post by the CEO.
Its a little surprising no one's done it so far, so I'll open the floor. What does the community think of the latest post, and do we trust (pun intended) the direction the company is taking?
Some thoughts:
At a macro level, we’ve moved well past the inflection point of the AI hype cycle.
Citation needed. This hype started 2 years ago. It is eerie similar to the Internet hype in the late 1990s/early 2000s which lasted some 3-4 years and didn't reach some apex from where it started to calm down. Rather it got more and more irrationally feverish with investors blindly investing in things they had no clue about, until the inflated hype ultimately burst in 2002, taking the whole software industry and the general economy down with it.
Was Internet not there to stay? Yes it was, and everyone behind the hype said as much back then. Don't conflate the hype with the core technology - just because some technology is here to stay doesn't mean that it isn't hyped. Far more likely it is hyped for that very reason.
These results highlight the fact that, while AI is here to stay, the quality of the results that AI tools produce is still very much in question by technologists.
The most obvious explanation for that is because those who work with the actual technology are also those most competent and best equipped to see that the technology does in fact not at all deliver what everyone claims that it does.
It is not so much a trust issue as a bad product quality issue. It could be as simple as people not trusting it because it is bad. As has been the case with early prototypes of every single technical invention throughout the history of mankind. Like for example the Internet in the 1990s.
For example, if you’re a backend engineer at a major financial services institution, are you willing to stake your professional reputation on AI-generated code without human review? Most corporate customers are not willing to take that bet.
Again it is not a trust issue. The reason is rather because there exist no tools which can generate such code. Most corporations are not willing to use tools that do not exist. They might be happy to sell you tools that do not deliver what was promised though.
With more than 58 million questions and answers across our site, Stack Overflow is uniquely positioned to meet this moment with our wealth of quality, accurate technical knowledge and the most trustworthy method to generate new knowledge going forward.
It is very naive to think that existing GenAI has not already scraped all of that out of SO, legally or otherwise.
How will the proliferating number of large language models continue to improve if there are no new human generated sources of content for them to train on?
Step 1 would be to get them working with the content they were already trained on. For creating amusing chat bots or pretty pictures, GenAI is a mature product. For generating source code, it is nowhere near anything useful. If it can't do that on material from year 2022 then it won't be able to that on material from year 2024 either.
And of course, what do we do when we commonly use these AI tools but can’t fully trust the information they present to us?
We don't use them. Perhaps we will in a distant future, who can tell.
Meeting developers where they are
They are here https://stackoverflow.com/.
In particular, feel free to interact with them here: https://meta.stackoverflow.com/
Or is there anything wrong with these sites? Are these bad products to avoid? I don't have the inside info why they are such bad products, but apparently this is the case and we shouldn't go there hoping to find developers.
At Stack Overflow, one of our core values is ‘Keep Community At Our Center’, and we take that ethos to heart.
I don't think there exists any community, as such. Perhaps there was some embryo of that back in 2014 somewhere. Nor do I think this hypothetical community is kept at the center of anything. Any sources or indications suggesting otherwise?
At Stack Overflow, we’re committed to our community, whether that be our public platform users, our customers, partners, and Stackers, but we also deeply believe in building a future where our larger knowledge communities can thrive in the era of AI.
Everything we've seen during the past 2 years indicates that a community, if such a hypothetical group even exists, prefers to thrive in an area completely free from the AI hype.
The ONLY thing AI has done for the Stack Exchange network so far is to provide floods of low quality spam-like posts and massive moderator workload.
Reading through the blog, I feel like there's a few common themes here. Critically, that people don't trust genAI.
At the end of the day, SE's not going to be a genAI company. Both teams and the public network are very much social software designed to help with human knowledge management.
One of the most striking takeaways for us was the fact that the gap between the use of AI and trust in its output continues to widen. In 2024 76% of all respondents use or plan to use AI tools, up from 70% in 2023.
Depending on how you interpret it, this might reflect a gap between decision makers, and the people using the tools. It might also mean better availability of such tools, as the FAAANG and other are putting in resources and trying to grab market share. It may not reflect a fitness of purpose for such tools for the tasks someone uses SE for.
At the same time, AI’s favorability rating decreased from 77% last year to 72% in our latest survey. Additionally, only 43% of developers say that they trust the accuracy of AI tools, and 31% of developers remain skeptical.
Its been a year - and one would assume that least some of change is due to people trying AI and finding it isn't actually as useful that one would think.
Anecdotally, I constantly hear this in my conversations with our customers. If developers don’t completely trust the output of the AI tools they use, the whole foundation of the house you’re building is shaky
In a well functioning network site - nearly everything we do is reviewed, questioned and otherwise tested. This question's comments reflects that. Broadly though, that people discuss, review and work through problems is essential, and something AI as designed is bad at. It could be a rubber duck, sure but
For example, if you’re a backend engineer at a major financial services institution, are you willing to stake your professional reputation on AI-generated code without human review? Most corporate customers are not willing to take that bet.
There's an old adage that the customer is always right. If nothing else, they always control your revenue. The problem might not be 'customers need to trust AI' - its that 'AI is not the right hammer to turn this bolt'
There's an old IBM slide making the rounds that reads
""A COMPUTER CAN NEVER BE HELD ACCOUNTABLE
THEREFORE A COMPUTER MUST NEVER MAKE A MANAGEMENT DECISION"
Software development, system administration and many of the other aspects of knowledge/tech work is about accountability. AI can't really be accountable.
I'd also say that many LLM/GenAI thought leaders see this as a way to shift costs from labour to compute - and well, if we replace our fledgling coders with AI, on the whole its going to be worse for us as a network.
Taking all this into account, what does the company leadership thinks it knows that the community and customers do not?
I'd keep this in mind for later.
This is why validated sources of data are crucial to ensure that the foundation is secure, and why we keep returning to how we can bridge this trust gap.
The thing is these sources of data exist - there's an entire ecosystem of tools for it which are well proven. I mean, SO for Teams is that. SO Inc isn't an AI company - it writes social software and I feel like that's a frame of reference that's often forgotten.
The community's output is the product, and the fact that its the 'human' web, as opposed to AI is its value. I'm unsure why bridging the trust gap of AI is in SE's interest. Clean non AI tainted data is somewhat trustworthy, as is the system of governance and quality control we have. This feels like an argument for having less not more AI in how we interact with the network.
I'll skip the parts about partnerships for now, and come down to the stuff that probably interests us.
At Stack Overflow, one of our core values is ‘Keep Community At Our Center’, and we take that ethos to heart. Our Product team here at Stack is constantly discussing ways we can give back to our global community after the trust and time they’ve put into contributing to our public platform for 16 years.
And yet - many community requests across many products and fields often end up mouldering or ignored. I'm pretty sure this and other posts won't get a response. There's a certain sense of weariness and burnout amongst the community. There's some positive things sure but a good many core complaints feel unaddressed and tech depth tends to pile up whenever the company shifts focus.
That's to say, actions not words.
I'd suggest considering the social contract that's gotten us where we are, and that perhaps as much as discussing, picking a few big asks, and making them happen would be nice. About a year ago, before the last round of layoffs, you talked about putting in 10% of the company's resources into genAI products.
I'd wonder if there was a payoff from this, and if some of these resources would have been better spent in more 'traditional' customer and community happiness projects.
Also as a skeptic that AI's a trend, not a bubble, and believing that the AI inflection is downwards - I'd ask again what's plan B should it burst?
To that end, we brought the hosting of our quarterly data dump in-house. You can view the data dump for each network site through the user profile. The idea behind this is that these quarterly data dumps are a crucial resource for our community, but we want to put safeguards around who is allowed to download the technical content on our site, and incentivize existing and potential partners to invest back into our community and join us in our mission of building the era of socially responsible AI.
Lets be honest here. Its a way to preserve a revenue source and try to keep the AI companies data kleptomania at bay. On the other hand - there's deep unhappiness with core community members, and in the year between the first, deeply unpopular attempt at this, and the current one, many of the concerns the community had - that the dumps were a hedge against the Stack Exchange Networks's parent organisation(s) being unsustainable or hostile remain unaddressed. I've probably waited, and reminded staff for over 3 months about a request for a full data dump. They're still working things out but - if someone familiar with the network and its processes has trouble, I wonder how a random academic without deep SE knowledge would do it. It would probably be faster to grab it off IA, or other means, but I'd like to do it the right way. I still can't get the old functionality, for archival reasons as such. I'm a trusted community member, and I can't do this the 'right' way.
I'd say that this is a perfect example of how trying to deal with abuse without considering 'honest' users can hurt community trust. This kind of is a pattern and hurts trust.
