Stackoverflow is investing into baking GenAI

I feel like GenAI has become the next &quot;blockchain&quot;. Wall Street (or someone else!) is setting some weird expectation for every company to somehow say &quot;we&#x27;re doing something with GenAI.&quot; Doesn&#x27;t matter what it is they&#x27;re doing with it, they just have to say the magic word in a press release or blog post.<p>Apparently, &quot;we&#x27;re using AI&quot; is a better story than &quot;we&#x27;re making great products, regardless of the underlying tech.&quot;

Take a look at the answers on this post if you want to know how their most recent genAI experiment went:<p><a href="https:&#x2F;&#x2F;meta.stackoverflow.com&#x2F;questions&#x2F;425162&#x2F;we-are-seeking-functional-feedback-for-the-formatting-assistant" rel="nofollow noreferrer">https:&#x2F;&#x2F;meta.stackoverflow.com&#x2F;questions&#x2F;425162&#x2F;we-are-seeki...</a><p>In short, it was a complete disaster. They create a tool to provide formatting suggestions on your new questions by piping the entire question into ChatGPT (or one of the comparable services). The internal prompt was figured out within minutes. And there are plenty of examples of this tool messing up questions entirely. And even when it doesn&#x27;t destroy the questions, it still edits far too much minor and irrelevant stuff.<p>I don&#x27;t understand how this experiment could make it this far, they actually put it into a live test on the site. So a part of the new users asking questions got this tool in this broken state. And the way this failed was entirely predictable when you just pipe content into genAI.

Can someone ELI5 for me what GenAI is? I&#x27;ve never heard of them and their website is just the usual content-free buzzword salad:<p><i>artificial intelligence</i> - what else, sigh!<p><i>transformative</i> - you bet<p><i>business AND consumer</i> - so basically everyone, why tell at all?<p><i>various industries</i> - everyone again<p><i>pioneering</i> - yeah, what else<p><i>vertically integrated</i> - sure, I was just waiting for that<p><i>AI solution</i> - oh, who would have thought?<p><i>AI-powered</i>!- you are saying!<p>I&#x27;ve literally not even finished the second sentence on the page and this rambling abomination goes on and on and on...<p>EDIT<p>For reference the first two sentences:<p><i>&quot;GenAI, our mission is to harness the power of artificial intelligence to create transformative tools that benefit businesses and consumers across various industries. GenAI is a pioneering artificial intelligence company focused on developing a vertically integrated AI solutions business through its proprietary MAI Cloud™ database, with the development and commercialization of AI-powered tools and solutions for businesses and consumers across multiple industries.&quot;</i><p>EDIT<p>After reading other comments I think GenAI is a generic term (like blockchain) referring to Generative AI. In my defense the OP is not enlightening in this regard at all, so I just googled the term and the company website of genai-solutions.com was the first result. Nothing in the OP says that Stackoverflow is working with this specific company, however, that was just my (likely erroneous) assumption.

Me: Hey AI how do I do this using X?<p>AI: Why would you ever want to do that using X? You should do Y.<p>Me: Well I need to do it using X because of these reasons.<p>AI: ...That&#x27;s dumb. You should just do Y.

Their traffic is rumoured to be down 25-30% after ChatGPT was released, so good to see that they are doing something to keep the community alive in the new era of GenAI.

They tried A&#x2F;B testing the “question formatting assistant” on the production site a few days ago. The results were…not great, and they had to quickly shut off the experiment: <a href="https:&#x2F;&#x2F;meta.stackoverflow.com&#x2F;questions&#x2F;425162&#x2F;we-are-seeking-functional-feedback-for-the-formatting-assistant" rel="nofollow noreferrer">https:&#x2F;&#x2F;meta.stackoverflow.com&#x2F;questions&#x2F;425162&#x2F;we-are-seeki...</a>

As writing, all comments seems to be quite negative. I think this is awesome and definitely something that can unlock a lot for me.<p>I problem I often encounter when using stack overflow is that questions are answered ina a very narrow way that doesn&#x27;t really fit me. So I need to bounce between 3-4 answers that are related to get the right solution.<p>I would guess this solution could do this for me, provided that I provide anough context is my problem prompt.

SO could go on all in using the 100k context length of Anthropic Claude, feed in the documentation of a library or product, and have it generate questions and answers, and post them on Stack Overflow to fake activity now that everyone I know participates on it in a read-only fashion.

The projects listed [1] seem a bit mundane at first, but I prefer that to the opposite, where companies shoehorn GPT into places where it just irritates users.<p>[1] Title generation, Formatting assistant, Sentiment analysis, Identifying duplicate questions

Duplicate question closing will now be instant if your question has an embedding that&#x27;s remotely close to another question

Just like wealth redistribution and inequality, I&#x27;m seeing this migration of &quot;programming discussion&quot; moving from open forums like SO to closed systems (guarded silos of GH Copilot, ChatGPT, Phind and others). We are preferring to ask the nitpicks in their chatbox instead of as a public comment on Stackoverflow.

In a way SO is the perfect platform for human feedback in reinforcement learning

The Stackoverflow research initiatives per generative AI are very good examples of what some enterprises can do to improve user-input text context on websites and applications.<p>1. Adapt user text to a standard format. 2. Correct spelling and grammar mistakes. 3. Detect inconsistencies during the input process.

Embedding search in a chat UI over all the SO content (with valid citations) would be the obvious feature I’d build if I were them.<p>“Me: How do I do X? SO: here are a few highly voted answers … Me: not quite, what if to foobar is like this? SO: In that case this is probably the answer: … Me: Upvote answer (giving RLHF to the system to help with future training)”<p>ChatGPT is great at answering code questions but if you are on SO you want to engage with the community, upvote the answer, or comment etc.

Whatever happened with the moderator strike over the AI? This seems like a great way to cheapen the brand.

Is this why ChatGPT answers are banned on SO? Because it would compete with their own upcoming business?

And before long it&#x27;s going to be, &quot;Let AI write an answer to a question and you only need to edit it!&quot;.