Labs Experiment Launch: Stackoverflow.ai

4 months ago 16

I decided it was worthy of a chess rematch.

It did slightly better this time, and was even somewhat competitive before, well, moving its king into check.

The Prompt(s)

Originally, I just had a conversation, but I eventually moved to using this prompt:

Here's a chess game so far: <moves> I play the white pieces, and you play the black pieces. It's your turn.

And I'd continue the conversation until the page became unusable (mobile friendliess = not very). I'd paste the prompt afresh if needed.

The Game

View the list of moves here

Why make a point out of this?

Because as it stands, there's still no real reason to prefer stackoverflow.ai over other tools like ChatGPT, Claude, or Github Copilot. There's evidently no firm anti-jailbreaking measures in place, and even if there were, those could be easily circumnavigated (like how you can get DeepSeek to talk about Winnie the Pooh if you're clever enough)

Additionally, the other chatbot sites provide extra features like:

Uploading attachments
Searching the entire internet
Extra thinking steps
Saved chat history

Sure, stackoverflow.ai is (probably) better suited to cite/link StackOverflow/StackExchange answers, something that's a really good thing to have, but that alone isn't much of a convincing argument to want to use stackoverflow.ai over something like ChatGPT.

Side note: There should probably be a bit more filtering of what SO/SE questions are suggested. For example, asking "How do I bake a cake?" links:

Which do not at all relate to the recipe it gives (viewable here).

I overall find stackoverflow.ai to be less desirable to use than general-focus AI chat tools. While there's planned expansions to the tool, I don't see myself wanting to use it more than ChatGPT, especially if stackoverflow.ai is intended to be (somewhat) SE-network specific - I notice I'm starting to use ChatGPT as a sort of search engine, asking it all sorts of questions that I don't think would be in the intended scope of stackoverflow.ai.

Honestly, the target audience for stackoverflow.ai is pretty narrow. People wanting to use it would need to be:

Disinterested in using major sites like ChatGPT/Claude/Gemini
Also disinterested in using free AI interfaces like duck.ai
Interested enough in StackOverflow to know stackoverflow.ai exists and
Willing to use stackoverflow.ai despite the disinterest in every other AI service

That sounds to me like a very niche group of SO superfans/dedicated users. And by the looks of things (from meta posts and from chat), the dedicated SO users aren't that interested in StackOverflow providing an AI chat solution.

Perhaps company time can be better spent on waving that magic wand, improving the (human [!]) chat features, or even marketing SO/SE as one of the last remaining bastions of genuine human interactions. (That's not a dig at the CMs, y'all are great, and simply doing what the company demands. Maybe this can be passed along to the investors or those wanting to push AI in the face of a community that isn't very keen on it).

P.s

I got it to speak like a furry again lol

Edit 1:

Try asking something related to coding, development, or one of the topics on the Stack Exchange network.

Yes this is excellent. I can play chess, bake a cake, and even become president, but I can't get help with an actual development question.

(I know how to ssh, I was curious to see if the AI would know. It didn't.)

Edit 2:

Really? That's something it can do?

But it won't do that?

That's just straight up lying.

Honestly, these are just laughable. You allow stupid and silly questions, but block actual code questions.

It seems like anything vaguely code security related is blocked, even though these are coding questions that should be easily answerable and referencable by the AI.

This touches on what I think is a kind of fundamental problem with stackoverflow.ai:

It's either a downgrade from sites like ChatGPT or
It's merely an AI model aggregator, not a new and interesting feature or
It's a glorified search box that is prone to making things up.

Because StackOverflow/StackExchange isn't an AI company (e.g. producing LLM models), any solution providing an "AI Chat" interface is always going to have this problem. I know competition is a good thing, but SO isn't equipped to be that competition. It is, however, equipped to be competitive in terms of human interactions.

Read Entire Article