Timeline for answer to Join us for our first community-wide AMA (Ask Me Anything) with Stack Overflow’s CEO (and a few others) on February 26, 2025 by Damir Tenishev
Current License: CC BY-SA 4.0
Post Revisions
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 26, 2025 at 19:43 | comment | added | Damir Tenishev | @KevinB, now I finally got the point. Thank you. Then, only scripts could be of use in case they have some tuning. | |
| Feb 26, 2025 at 18:48 | comment | added | user400654 | I mean... if the LLM they're using to do summaries is just sending stuff off to gemini... i'm not sure what you're expecting to retrieve. | |
| Feb 26, 2025 at 18:47 | comment | added | Damir Tenishev | @KevinB, as a second thought, even this could be used locally. If they share LLM which is doing summaries and allow to use it with prompting, it could work. Like I have command prompt tool where I post my question, it sends it to SO/SE, gets the results and builds the answer. | |
| Feb 26, 2025 at 18:46 | comment | added | user400654 | Honestly I wasn't sure i completely understood what you were stating, if maybe you were interpreting something from what i said that i didn't think i said or... what. :shrug:. but it was more or less just semantic search + summarization, and a fairly standard chatbot mode. From what i've seen of demos for what it in teams... it seems like it's exactly the same thing. | |
| Feb 26, 2025 at 18:45 | comment | added | Damir Tenishev | @KevinB, I get it, thank you. (Just to be on the same page, in my last comment I just asked if I got you correctly, this wasn't a kind of argument or doubts in your vision for this case). | |
| Feb 26, 2025 at 18:41 | comment | added | user400654 | My understanding, based on the feature announcement for when this feature was publicly tested on stackoverflow.com, was that it was simply a search + summarization. no training. They relied on existing public models. | |
| Feb 26, 2025 at 18:39 | comment | added | Damir Tenishev | @KevinB, do you mean that they use real SO/SE textbase and real-time search on it without any preprocessing which results in a relatively small dataset (like trained 8Gb LLM)? Do you see like they just run search with my text and then compile results only with LLM? | |
| Feb 26, 2025 at 16:36 | comment | added | user400654 | my understanding is... they don't have a trained LLM for this purpose, they're using off the shelf solutions, combining a search that finds relevant answers and an LLM doing summaries. | |
| Feb 26, 2025 at 16:25 | comment | added | Damir Tenishev | @KevinB, thank you, this helps. Seems, the question should be reworded to "does this mean that this won't be on the other parts of SO/SE, since it is a valuable part of paid section and a good reason for pay?". Not sure who could I ask about this. If this is the reasoning, there is no hope to get the trained LLM. | |
| Feb 26, 2025 at 15:35 | comment | added | user400654 | This currently exists as a stack over flow for teams feature. | |
| Feb 24, 2025 at 10:53 | history | answered | Damir Tenishev | CC BY-SA 4.0 |