1y ago

Stack Overflow Just Announced Their Own AI OverflowAI

Let’s highlight the new features and products we announced today from the stage of WeAreDevelopers.

OverflowAI

11 4

186 72

25 comments

User: OverflowAI how do I tie my shoes?
OverflowAI: That's a dumb question. Just wear velcro shoes.
User: But the thing is I really need to know how to tie my shoe laces.
Stack Overflow Mods: [This post has been answered and is now closed.]
- OverflowAI: duplicate of “how to tie-dye a shirt”

As an AI model, I cannot answer duplicate questions. Please check this completely unrelated question instead.
- As a language model, did you even RTFM?
  
  Alternatively: as a language model, found the issue!
- No even an unrelated question, just a completely made up url.

Great, now an AI is gonna tell me my questions are dumb
- Easy, just use AI to word it for you

So it's just going to automatically tell people (in the rudest way possible) their question is a duplicate and kill the thread for each and every new post.

Will the ai be as cowardly as the people who vote to close without comment?
- Better than the ones with 300k reputation who comment to tell you why your question is shit and encourage everyone to close while making up or grossly misinterpreting the rules.
  Then they find you a few weeks later and do it again.

Oh great. Another place on the web where my usage of it will simply feed some new language model.

Just yesterday the news was that their site traffic plummeted. The solution? Throw AI at it

That's can be nice !
Until now we’ve relied on lexical search, trying to match users with questions and answers based on the keywords they supplied. But as I announced today, we’ll be adding semantic search in a private Alpha, built on top of a vector database, so that the responses generated from a search query can more intelligently align with the topics the user is researching.
Sounds like we don't need to know the right word to discover the good response. 👏

The generally unhelpful attitude of the community aside, I don't think these one-of-a-kind problems lend themselves to AI very well.
- What makes you think that? Just yesterday I asked "how can I make an excel file in X language?"
  GPT-4 suggested two libraries, one popular but not maintained, the other new(ish) and maintained. It told me how to install the latter and provided sample code for a simple hello world spreadsheet.
  I was then able to ask for a few tweaks, like sending output to stdout instead of to a file on disk, and using 4,2 cell coordinates instead of D2 coordinates. It was able to tell me how to tell Excel that "4 Feb 2042" is a date value, not a string value, make the heading columns bold, make the column widths automatically calculated based on the content, etc etc.
  The whole think was so one-of-a-kind I doubt you could find questions to all those answers (especially with a specific library) on the regular Stack Overflow and even if they are there it'd be a dozen separate questions that I'd need to piece together.

25 comments