@KevinB In principle, we have enough expertise. In practice, we don't have any way for people to describe that information (duplicates are a really poor alternative), so that knowledge only lives inside people's heads. (LLMs are a really crude, lossy way to extract this information, and the fact they're so popular validates the need for something like this.)
I've had designs for such a system knocking around for years, but I wouldn't be able to bootstrap its database on my own (and tbh I'd want to re-do those designs, now that I've read of more prehistoric academics who almost produced the system I'm describing, but didn't quite get there).
People like the Rust and Haskell compilers. There's no inherent reason that a tool like that can't solve all problems, with nice friendly descriptions written by real people who've faced this exact problem before. "Help" questions could have the reference components and the holistic "how do we get to the canonical references from this instance of the problem?" information extracted, put in a form that generalises…
> There are many Python questions that a fairly simple lint tool could automatically answer, even before they're asked. (Sadly, I lack the attention span for making such a tool under my own steam. I've tried, I've probably still got my notes: it just didn't happen. I probably could do it if I were salaried for it.)
@Slate Visualization is an important part of manifesting ;)
@Slate I'm now mostly left with questions like ... How is e.g. that leaked Answer Assistant going to do some of that? But that's a different discussion better kept out of here for now, I guess :)
Part of that research seems to suggest that the AI/LLM is somehow wanted as 'part' of the community, with community editing/approving/vetting their posts as if they were just another user. I now wonder if that's ever going to happen... maybe if it gets really good, but with the quality of things LLMs/AI spew out these days, I'd probably mod them like any other user and would want to see them answer-banned :P
@Tinkeringbell I wonder if that lack of "accountability" is part of why people (myself included) find GenAI so offputting... Like, a human hates to write bad answers. They get social pressure to either be better or just not post at all on a topic they can't really answer about. For most people, that's plenty; if anything, the balance is probably towards scaring some people away from answering at all. The people who stubbornly keep at it are a rare few that get handled by system restrictions and mods
But for the AI... the AI doesn't care. Someone posting AI content doesn't really care, either; it's not their own work being judged, even if they're the one earning rep or not. That social pressure is missing
@Tinkeringbell it's a fair question. not sure I have a good answer for you atm. there's a part of me that wants to say struggles with AI is directly relevant and a part of me that wants to say it's little more than a case example of something deeper
@BryanKrause yeah, I read those! they're definitely relevant
I think part of what frustrates me about GPT output, where i see it most often in search, is that it's more or less being presented instead of the much better results we used to get 5+ years ago
but there's likely no going back to that, given how flooded things are with garbage now days
so.... :shrug:? guess it's an acceptable compromise in some regards?
If I care heavily about the accuracy of the answer, like it has relevance to the health of my kid or is important for winning an argument about trivia, it puts me squarely back where I started though which is finding some reference that is trustworthy
looking for documentation, methods that exist, etc
from that perspective any hint is enough to lead you to proper documentation
It's certainly frustrating that i'm getting a summary instead of a link to what contains the info i need, but it's better than nothing at all
effectivley what has happened is instead of providing me what i want through generic search result listings like they did in the past, they're just not being provided in leu of showing a summary of them instead tha tyou have to click through to reach. the frustrating part is that i need to click through this summary to reach it
One positive though is it'll summarize content from some of these clickbait ad driven garbage dev sites so i can benefit from their content without giving them a view,
It's all worse than what i actually want, a link to a source so i can draw my own conclusions
I find SO most useful because it often comes with examples that are sparse in documentation and I often find it a lot faster and intuitive to learn from an example rather than reading the docs, and then use the docs to elaborate
In a decidedly un-American move, I'm not going to be working over the holidays.
In fact I'm leaving end of day today, and I won't be back until around mid-January.
4
So if anyone blithely accuses this post of being "just another broken promise," another "dead end discussion that went nowhere," etc., I'd politely request they be informed that I am entitled to my vacation ;)
yeah, I hear that. I mean, I do think it's overstated to a degree, but it's impressions and opinions all the way down
for me, in a way, writing is thinking. and that makes it easy (if a bit labor-intensive) to communicate that I'm engaging with something. provided I have the latitude to talk about it openly etc etc lol
Yeah I'd appreciate if Philippe gave some indication that he read the responses there (and in other posts), if not giving some responses where he can. I know it's quite likely everything gets read, it just feels like nothing got read because he was radio silent.
fwiw... I get what you're saying, and also, I don't really hold it against him. (well, obviously I'm biased, since he's my boss.) functionally speaking, I've gotten nothing done the last three days except have this conversation
which is valuable, don't get me wrong. but he's a manager. he's got people who depend on him within the org. shit happens, sometimes the couple days one wants to take responding get eaten up by something else, and by then the question's cold
also like, gotta take into account that people hated that question, for one reason or another. that makes responding harder (and more time-consuming), too
Yeah I mean starball was pretty ruthless here (no offense to starball, they're great, it's just that's... A pretty rough way to just lay it all out like that)
Oh, hey, look
Woah, this is just brutal. Why post this? It seems like an effective way to get Stack Exchange to not want to genuinely engage with us. — Rebecca J. StonesNov 18, 2023 at 23:48
oh yeah if I were in his shoes, I don't know how I'd even begin approaching that post (sorry starball lol)
it lays out the terms clearly, but the terms are not really favorable for honest discussion. if I got that answer my strategy would probably be "ok, wrong place and wrong time. point taken. I'm going to scrub this attempt and try again later in a different way"
(it's something I was a bit afraid would happen with the new post too, but I didn't really think I'd get a better opportunity by changing the timing. esp going on vacation. january is just too late to start folks thinking about heavy questions)
lol, not really intending to hint at something specific with that, sorry
I guess more generally speaking, the fact that I'm posting this should reasonably tell you I'm not the only one who thinks the time is right to ask serious questions about the big picture. maybe balpha's answer also suggests it's on many people's minds, too. I put together this post, but I didn't spend this time on a lark
my point was more that it's kinda been the status quo to post things like this then kinda... leave it behind for the past few years. It's nice seeing it be handled differently this time
