General Tildes feedback, questions, and so on
Things have been a little on the quiet and steady side for the last while now (which has been nice), and it's been some time since we had a thread for general questions and feedback. Feel free to use this thread to post things about the site that you're curious about, questions or suggestions you have, and so on.
@super_james was also nice enough to start this other thread today about ways to help, so this one will probably be a little more on the "specific things to help with" side.
As a couple other things from my end, I've topped everyone back up to 5 invite codes again, so you should have some available on https://tildes.net/invite if there's anyone you'd like to invite to the site (and as always, just message me if you ever need more codes).
I also posted some information the other day about donations so far and the general financial status of the site, so that might be something that you'd be interested in reading if you didn't see it already. That thread also seems to have motivated a decent number of other people to pledge to the site's Patreon or donate, so thanks to everyone that donated, I really appreciate it.
As a final thing, a couple people have asked me about the site's general activity levels lately, so I'll post some stats about that in a comment here in a little bit (so it's more easily collapsible than being in the post itself).
As always, thanks for being here!
I had my first experience receiving an "Exemplary" label this week. I know it's not a super new feature, but I think it's a great "super upvote" thing I've always wanted to be able to give for outstanding contributions. The one per 8-hour limit seems like a good choice.
Related to that, I wanted to ask:
I thought that it was a cool idea when it was introduced, but it wasn't until I got a couple on a post that I really realized the value of the exemplary label. It felt good to have the work I'd put into the post recognized, and the addition of the "reason" was a big part of that.
It's probably silly of me, but it made me want to contribute more and with more thought. (Not necessarily to rack up more exemplary labels, but because it reinforces the idea that those sorts of posts are valued by the community and worth writing, even if you don't get a direct reply.) I'm guessing that's the intention, so I can tell my brain "good job" for being predictable, haha.
I also like that there's no running count of anything, not the number of votes you've received, nor your number of
Exemplarylabels. There's no always-visible karma display like on reddit, so I'm not constantly worried about whether a post is going to be popular in the "gets a lot of votes" kind of way—not that I was worried about it after 12 years, but still. Sure, we can keep a general mental image of whether our comments "do well" here, but that's it, and I'm liking it.I've not used Reddit regularly, so I only really got introduced to the concept of karma during early meta discussions here on Tildes (many of which delved into why it wasn't a feature that would be implemented). It does sound like it could become a bit stressful, or reward unwanted behaviors.
I do find myself paying attention to vote tallies for my individual comments on Tildes, but it's more of a curiosity thing, I think. "Oh, neat, this resonated with 14 people in some way." I guess sometimes if a comment doesn't do "well" I can feel disappointed, but... I don't know. It doesn't feel particularly bad, not in the same way that I can imagine being downvoted to hell can feel.
Same. Even though I sometimes like accumulating points on sites like Stack Overflow, the fact that there is no karma/reputation/trust score total visible anywhere means that Tildes is less likely to fall prey to Goodhart's Law where people seek points for their own sake and the metric becomes the target.
I need a master list of these internet laws to crib ideas from. :)
On Reddit the upvotes and downvotes are absolutely "agree" or "disagree" buttons in practice. Voting on Tildes, while not absolutely immune from similar problems, feels much more like a "you've made an valuable addition to the discussion" button.
I think they're mostly used as agree/disagree. You can always tell which comment will have more votes by the popularity of that opinion amongst the site's demographics.
I think that tends to be correct for the top 1% (pulled out-of-my ass number) of posts in terms of comment activity. The posts generating 150+ responses tend to be controversial and/or political and the comments start to feel more like Reddit in general.
So if your comment doesn't get any votes, does that mean people disagree with it?
... but wouldn't that make you right?
Argh
I tend to cop a lot of flak and encounter a lot of disagreement here. It's my nature to be contrarian, and that gets reflected back at me. It was therefore very pleasing to have one of my recent comments labelled as 'Exemplary' - and not just by one person, but by quite a few people. That was very affirming. I couldn't help having a brief Sally Field moment in response to that. :)
I've had a couple of other comments receive 'Exemplary' labels (but only one each!), and it's always good to receive that extra little bit of positive feedback.
The trust system is still mostly just a vague concept, since I don't think it should be very necessary until the site gets much larger. It's something I keep in mind when I'm planning things, and there are some general pieces built that will be useful for it eventually, but there's not much concrete yet.
As an example, the comment-labeling all works based on "label weight", and when a user adds a label to a comment it adds a specific amount of weight to that label. For now, everyone's weight is exactly the same so it's not any more meaningful than a simple count would be, but the ability is there to be able to give some users' labels more or less weight, and that's the sort of thing we'd want to be adjusting with the trust system in the future.
Is there a link to details about the trust system? I read the mechanics, but that didn't seem to discuss it as far as I could tell. The entire reason I joined Tildes is because I've been waiting for a Reddit clone to do this.
As far as suggestions go, I have two of importance:
Please keep the invite-only, or at the very least keep a "vouch" option for users, so that people are linked to each other for the purposes of trust. If the site becomes popular, propagandists will come. And it will be nigh-impossible for them to maintain a network of untrustworthy people if their trustworthiness drops, and as a result the other people they vouched for lose voting power.
There really needs to be a Fact-Checking option, or something that can clearly be marked as non-factual, or citations needed, and the post/comment should be appropriately adjusted based on this. Because a lot of trusted people DO fact-check, and their efforts should be reflected by that.
The general idea of the trust system is explained here: https://docs.tildes.net/mechanics-future
I'll keep my question pretty open ended and easy to answer.
Are there any features nearing completion or ready to put on tildes in the near future? Just want to get an idea of what types of things I should look out for in the coming weeks.
As the others mentioned, the "Roadmap" link in the site footer goes to this page, which is usually a decent overview of what's being worked on at the moment, or should be getting worked on soon: https://gitlab.com/tildes/tildes/boards
Outside of what's shown there, I've been trying to gradually work towards some things that will be needed to make the site publicly-visible so that we can start allowing people to browse it without needing to get an invite first. This mostly revolves around some features related to how groups work and supporting "browsing" them more easily, because I don't want Tildes to have a "default front page". I think that caused a huge amount of issues for reddit over the long term, so I'd like to make it so that viewers have to be more deliberate about what they want to see.
Would that be a sort of wizard that asks what a reader's interests are, and points them towards those groups? That's something I could see at signup, but how would the front page change for non-logged-in viewers? Through a cookie, perhaps?
I'm not totally certain where it will go yet, but initially I'm just intending to make logged-out users go into groups individually, with the front page just being an easy way to pick which one to browse. The idea of a "merged" front page can be nice, but really isn't essential. There are plenty of sites/forums/etc. where you have to pick a category, and it works perfectly fine. You usually just end up with people going directly to the categories that they care about, instead of using the front page at all.
Interesting. It makes sense, because it's in essence an aggregator, and some news sites for instance will ask your interests before presenting you the news. I certainly don't want to read about celebrity fashion, so it's great when they let me choose what to see and what not to see.
Other than merge requests, which you mentioned, the "Development" board on Gitlab is the easiest way for people to keep tabs on "In Progress" and "High Priority" issues/features IMO:
https://gitlab.com/tildes/tildes/boards
I’m slowly working on bookmarks, user bios, and a back to top button on mobile.
I just manually deploy it, it's not much of a process. Almost all of the time it's just: pull the code from GitLab, run Alembic migrations if necessary, restart webserver. It's possible that I'll move to more of a CI pipeline in the future, but it would be pretty overkill for now.
I wouldn't say that Tildes uses database triggers heavily. There are some, but they're mostly very basic and just handle simple data-integrity tasks that the application code really shouldn't need to handle manually.
As for the API, the way the code is currently set up, there's something referred to as the "web API", which is intended to be used by the site itself and mostly responds with HTML fragments that can be swapped into the pages. The "real" API would probably have a lot of similar endpoints, but will probably be a lot more formal (instead of being based around "whatever the site needs") and respond with JSON.
I don't understand your question about the relationship between the API and the database triggers. The API will just result in data being changed in the database the same way the site does, which will cause the relevant triggers to run when they're needed. It shouldn't need to be treated any differently.
Thanks for elaborating, I completely understand. A lot of people definitely use triggers very poorly, and I think that's made a lot of people scared of using them at all. I think that's a huge shame, because they're extremely powerful, fast, reliable, and can simplify a lot of things from the application's perspective.
My overall approach is to try to use triggers for things that are more related to data integrity than actual application logic. So for example, when someone deletes a comment, a trigger also deletes any notifications that were related to that comment. That's not something the application code should need to remember to do every time—there's no case where you want to delete a comment but leave the notification behind. So that's a perfect place to use a trigger, and you can be confident that you're never going to have weird "ghost" notifications left behind because of an error somewhere in the application code (or a request being terminated partway, etc.). The database makes sure that all the necessary updates happen together.
Is that what causes reddit to sometimes add a comment to someone's comment history but neglect to add it to the actual comment section "tree" where they tried to post said comment (e.g.)?
Not quite the same cause, but kind of a related class of issues. Reddit relies really heavily on updating a lot of different cached versions of listings (including comment trees) whenever something changes, and sometimes some of those updates fail. It's especially hard because they're using multiple different data stores, and keeping the data in sync between them is difficult.
Is there no easy way to verify uniformity across multiple caches/data stores (e.g. comparing hashes, double checking updates actually went through, etc)? Will this become a potential issue with Tildes as it scales up and you need to similarly start relying on multiple caches/data stores or is your data architecture fundamentally different than reddit's so it won't be an issue?
There's a famous saying:
Avoiding caching issues is a really hard problem for a lot of reasons. It's possible that we'll run into similar issues eventually, but it'll be a long way out.
Multi-master databases are still one of the things most companies have to go commercial to get. Postgres doesn't support it yet natively though there's a lot of interest in getting it to do that fairly soon. There are even several companies offering proprietary version of Postgres with this capability added. But fuck proprietary. Open source is always good enough.
There is an open-source fork of Postgres called Postgres-XL, based on v9.5 that is designed specifically to be multi-master. I haven't had the pleasure of tinkering with it yet myself, but the feature list is nice. If Tildes is looking for a multi-master toybox I'd start there. There are companies that offer commercial support for PG-XL too.
Edit: And they have betas for Postgres 10 as well. Looks like a healthy project.
How do you prefer to deal with horizontal scaling complexity?
how do you prevent downtime during the update? or is Tildes down for a few seconds while you grab the changes and do the restart?
Most of the time, there's no downtime. The web server just gets reloaded, which (at least as far as I know) results in it smoothly updating over to the new code for new requests.
About the only time there is any downtime is if there's some sort of breaking database change, like removing a column that the old code was using. Then there's usually just a couple of seconds of downtime while I update the database and switch to the new code.
The one I could use right now are custom sorts. That is, I want to be able to have "Most comments in the last 30d" and "Most votes in the last 7d" right next to the default sorts. Right now I can achieve that by using bookmarks or just typing a period as a query parameter into the URL, but I think you can see the problem(s) here.
(In the meantime, just adding "last 7 days" to the list of periods would suffice. I mean, who needs "last 3 days" really?)
Another feature that I often miss on Reddit is something like "top of the year" where I could see top posts from, say, 2012. HackerNews has a similar feature where you can see what the front page looked like on a certain day: https://news.ycombinator.com/front?day=2011-07-12.
Yeah, I think it'd be great to be able to set up your own custom "views" of different listings. Right now you can set the default, but that's about it. The site can handle it (which is why you can already choose "other period" and enter any time period you want), but we need a way to save and use them more easily.
And yeah, being able to set both a start and end date/time would enable things like that HN feature. That way you'd be able to say things like "show me the top-voted posts between 2011-07-12 and 2011-07-16", or whatever other combination you want.
@super_james has been doing some work on the open-source code that makes the ability to set time periods much more flexible, so that will probably end up being very useful for these sorts of things.
Sorting by specified date/time ranges is also already on Gitlab too, although nobody has started working on it yet AFAIK:
https://gitlab.com/tildes/tildes/issues/237
I actually really like "last 3 days". It's perfect for if I want to catch up on what happened over the weekend. Also, I'm hesitant to interact with stuff from 4-7 days old as it feels like "necro-ing" in many cases.
I see. I use the 7 days period on the week-end, when I binge the content I've missed during the work week.
I'm not sure what's the policy on necroposting here on Tildes is, but I think that a week-old post is probably still relevant. At least I've seen week-old posts re-emerging several times.
There is no necroposting policy AFAIK. And I personally think complaining about "necroposting" here is a bit absurd since, unlike traditional forums where necroposting can be an actual issue, on Tildes if a user doesn't want to see posts reappear beyond a certain age then they can just set their activity sort threshold to lower than that.
The only time I have seen topics show up on activity sort that I felt miffed at is because someone commented on their own 0 comment submission hours or even days after originally posting it. If it's an update to the topic, that's fine with me and perfectly acceptable... but I have seen a few times now where it was clearly just an OP purposely delaying making a comment to exploit the activity sort.
...eh, I've done that...delayed by hours anyways (not days).
Sometimes I intend to write something, get interrupted and have to stop, and I don't get a chance to get back to it for a few hours. Is it honestly something that's really annoying? I use to write everything up in notepad or something first and then copy it over, but honestly got lazy and stopped that, which is why there is a delay in my posting and my commenting sometimes.
It really depends on the comment, TBH. If it's a substantial contribution then a few hours delay doesn't really bother me, but OPs commenting on a >24hrs 0 comment topic is kind of pushing it no matter the circumstances IMO. And the ones I am referring to that were "clearly" exploiting the system were far from substantial.
I've done this quite a few times, and it's not because I'm trying to abuse the activity sort or deliberately delay my comment. It's usually because I post an article that I recognize is an interesting article, important piece of news, etc. but I haven't had time to actually read it myself yet. Then hours or days later when I get around to reading it is when I might have a comment about it.
Steal one of 4chan's more clever features - the 'sage' tag. Enough people flag a post that keeps popping up with that, and it stops bumping for a while (or forever as the number of sage tags piles up). That seems like a decent user-powered way to identify pointless bumping. Might even be a halfway decent honeypot to identify spammers and other agenda-driven accounts.
Fair enough, but I honestly still think on 0 comment topics that OPs should probably not be able to "bump" them with a comment of their own, especially hours/days later.
It really depends on the content of the thread. For general discussions not tied to any point in time, I don't have much issue adding a new comment to an old thread if I have something to contribute. Other stuff relating to news or current events can often become outdated as things rapidly evolve.
This comment goes into it: https://tildes.net/~tildes/7dz/what_sort_are_you_using_for_your_front_page#comment-2155
As does this comment: https://tildes.net/~tildes/7dz/what_sort_are_you_using_for_your_front_page#comment-215x
In short, reviving older threads is encouraged here.
Yes, all logs (of all types, including topic logs) are deleted after 30 days.
I don't object this approach, but I'm curious as to the line of reasoning involved. Storage/clutter related or more privacy oriented?
It's definitely privacy-related, but overall I just think it's harmful that we've gotten to a state where the default approach is "keep as much data as you can, forever, even if you're not using it for anything". That's what makes data breaches so scary—companies keep everything, even when they don't have a particular use for it. I think data should be treated more like a harmful byproduct, and only kept when you actually need it for something.
But, in this case, isn't the data relevant? For your proposed trust system, don't you need to keep a record of what sort of changes a user has been making? How do you know whether someone has been making good changes and should be promoted to higher duties, or has been making bad changes and should be demoted, if you don't retain that topic log data?
Yes, but it doesn't take over 30 days to make those determinations.
How would you track user behaviour over the long-term, though? How will you know when a user is changing tack from their previous behaviour?
You don't need to permanently keep a record of every single action a user took that resulted in trust +/- to determine long-term behavioral trends... the trust score itself does that for you as a meta-value. And just because you don't keep the record of every user action or the topic logs doesn't mean you won't keep all the records of any major infractions, commendations or significant shifts in user trust values, which can also be used to determine behavioral trends.
I think you're probably thinking about it too theoretically, because as a general concept it definitely seems like having as much history as possible would be useful. But in practice, we're never going to need months of history to figure out things like whether people edit titles responsibly.
Privacy is the cake, the performant benefits are the icing. ;)
What are topic logs?
It's not on every topic, but if the OP or a moderator edits a post it be on the sidebar
eg: https://tildes.net/~talk/7fa
I would like to have at least the interface in my language, so for me i18n is a priority. Maybe we can have discussions in other languages too at some time, but for now, just have the site translated would be nice. I'm sure a lot of tilders here would be happy to help with this.
So do you have plans about this ?
Overall, it's really not something that I'm working towards at all at this point. As mentioned in the other recent thread that people linked, the plans for the foreseeable future are to keep the site's content in English, and I don't think it's very meaningful to translate the interface if people aren't able to read the actual content anyway.
It's not something I'm opposed to, but I expect that it would actually be more useful for other people that might want to run their own different-language version of the Tildes code than for Tildes itself. Because of that, it's not really something that I'm going to prioritize anytime soon, there's too much other essential work that will have immediate benefits.
Relevant discussion: https://tildes.net/~tildes/6gr/are_we_ready_for_internationalization_i18n
This was basically discussed about a month ago. I think that eventually discussions in other languages should be supported, although I don't really care about the interface i18n.
Since the site's activity hasn't really been increasing much, I still don't want to fragment into too many groups yet. I've been doing some work related to the group system lately though (which is generally aimed towards being able to make the site publicly-visible), so there will probably be some group-related changes in the somewhat near future.
I was planning to have another thread to propose new groups pretty soon anyway though, probably early next week. Some of the existing ones aren't getting much activity (which isn't necessarily a bad thing), and there seem to be a few gaps that I think would be good to try to create groups for. Specifically, I think something along the lines of ~business would probably be a good idea, there seem to be quite a few topics that would fit that getting posted in ~life and ~misc lately.
Is there a particular group/subgroup that you'd really like to see?
The way I understand it from reading a lot of discussion about subgroups, that sub would be considered when there are enough mechanical keyboard topics in ~hobbies to justify it. Right now, that doesn't seem to be the case. I, for instance, would love a category for watches, but there isn't currently the content load here to justify it.
That's not the only mechanism for determining when a group would be added, there are others. E.g. A ~tildes.official topic asking for group suggestions was made ages ago, and Deimos has said another is likely coming some time next week as well.
Deimos has also previously talked about how having no group for a specific subject can have a chilling effect on the amount of submissions related to that subject, so determining interested in said subject purely by tags isn't necessarily all that accurate. E.g. ~anime wasn't created based on tags but on a suggestion and it's reasonably active despite there having been very few anime related submissions before the group was created (probably since being forced to submit things to ~misc can be discouraging).
I had actually wondered about that when I first went down the list of groups and saw anime there. It seemed particularly specific compared to the more general scope of most of the others. I'll have to make sure to suggest a couple I'd like to see in the next round.
Side note: Interestingly, your link took me to a "no topics found" page until I changed my sort to
all time.Hmm... that's not good. Linking to search results and tags should probably automatically ignore users' default sort period. Created a Gitlab issue for it.
I expect we'll be able to reshuffle positions of groups in the hierarchy. Sub-level could become top-level and vice versa, or groups could jump chains, or even be multi-homed in several places at once. Depends on what the community needs as it grows. Some amount of reshuffling will be natural as topics diversify.
That means there's really no harm in being generous with top-level group creation. We won't need to go micromanage-crazy until the hierarchy is much more complex than it is today, or likely will become in the near future.
I'm guessing this isn't an intended feature, but I've been trying this for a while to see how many I can generate - when you re-up invite codes, if a user generates the codes but doesn't share them, they can keep re-upping. Right now I'm at 28 unused codes from nothing special.
While I don't have 28 invites to send, it's possible something like this could be abused to mass-add accounts that would require some work on your part to remedy. If you didn't already know about this, now you know!
Yeah, I'm aware it's possible, I'm not too worried about it. You could have just asked me for 30 codes and I would have given them to you, so it's not really any different in the end. If I see someone actually abusing it somehow I'll worry about it, but I'm not really concerned.
Will there be an app or mobile/compact version of the site?
The existing site is responsive and works quite well on mobile, though some aspects do need more work. It should have an "add to home screen" type button as well if you're on a phone that can split it out into something a little more app-like (and there are also other options like Hermit that do an even better job of this).
The goal is not to have a separate, official "mobile version". It should always just be the site itself, which means that mobile and desktop users have access to all the same features at the same time.
And one on Reddit too:
https://www.reddit.com/r/tildes/comments/9blx0x/i_really_wish_there_was_a_mobile_app/
As well as one about the Progressive Web App (PWA) manifest:
https://www.reddit.com/r/tildes/comments/9cxfpj/pwa_support/
No worries... this topic is for asking questions and even if they are "repeats" it still gives more people an opportunity to see the answers to them as well. :)
p.s. It's also mentioned in the Docs, which are also well worth reading at some point too IMO:
https://docs.tildes.net/technical-goals#the-site-is-the-main-mobile-interface-not-an-app
there won't be, you can add the website to your home screen and obtain the same experience
Is there a way to do it on iOS? I just switched from Android, and it was dead-simple there, but I've tried Safari and Firefox on iOS, and neither seems to support that.
In Safari click the [↑] icon in the top right, then in the bottom row where "Copy, Add to Reading List, etc" is, scroll over until you see "Add to Home Screen" and select that.
p.s. If you want a more "app-like" experience and don't mind putting in a bit of work then I would suggest getting iCab mobile. It has a really nice, ultra-minimalist full-screen mode and all sorts of deep customization options, including UI and custom gesture behavior which you can set up to have Tildes look and behave very much like a purpose built app.
I can't believe I didn't see that. The [↑] icon is at the bottom middle on my iPhone Xs on iOS 12, but I didn't realize that list scrolled horizontally. Thanks for that. I had no problem getting used to iOS, but the little things are sometimes hidden in places strange to a longtime Android user.
iCab mobile looks pretty fantastic. I'll grab it when I have a bit more time to play around. For now, ~ works well in Safari.
Yeah, they really should do a better job of indicating those [↑] rows are actually scrollable since it's pretty unintuitive. Even just a simple > beside the right most icon would help people realize there are more options.
And yeah iCab mobile really is fantastic. I only started using it about 3 months ago when someone here suggested it to me, but it's my go-to iOS browser now.
Can you create gestures for navigation, like swiping right or left to go back or forward? I know mouse gestures never really took off on desktop, but I use them on every setup I have and hate using a computer that doesn't have them installed in the browser, so I'd love to just swipe for navigation.
Yeah you can set gestures and multi-touch swipes/taps for page navigation, reload, open/close tabs, bookmark management, switching to reader mode, open particular pages (e.g. your Tildes user settings, inbox/notifications or history), etc. There is a ton of options.
Thanks again. I had no idea that browser even existed, and I've been spending time searching top apps for iOS since switching. Definitely going to get it now. That sounds amazing, and I don't know of any other browser that supports that.
I think making the site publicly visible but still invite only will result in more users but still maintain the high quality of discussion. At the moment we've got a relatively high barrier for access (find site somehow, keep eye out for invites, get one) without being able to see what they're getting into. Large commitment in internet terms.
Pushing the site again on social media/reddit/hacker news etc etc when it becomes publicly visible is obviously a good idea.
Maybe there could be a simple metrics page that is automatically updated?
Finally just a general thanks for creating this place, it's fun to hangout here.
Two features I miss from HN which I think would make a good addition are submissions by domain (example) and "Previous submissions". The latter isn't so much of a HN feature, just a search query, but I think it encourages reigniting and/or referencing old discussions. I'd love to see that concept taken further, because it makes submission aggregators feel like less of an information black hole after a few days.
Yeah, I'd consider both of those features that would be related to search. Search generally needs a ton of fleshing out, the current implementation is extremely basic.
As tildes gets more popular, and comment trees grow, the practicality of the "Post Comment" box being at the bottom of the comment tree decreases. Possible solutions to this range from simply moving it above the tree, or having it as a static block in a corner of the page. Have you thought about this issue?
I disagree. As comment sections grow the value of new top-level comments decreases as the likelihood of them even being seen decreases as well. so providing an easier means to create them at that point is pretty pointless IMO.
Have you ever spent some time in large comment threads on reddit sorting by new? It's an endless stream of utterly pointless noise with very, very little substance and it's extremely rare to see any votes on said comments, even the rare substantive ones... And TBH I think that problem is an inherent property of new top-level comments in super large comment sections, since people who understand that once they reach a certain size making a new top-level comment is pointless don't make new top-level comments at that point, however people oblivious to this fact (who are the ones more likely to just create noise comments like "lol" and "this is dumb", etc) don't realize that and so are the only ones likely to continue making new top-level comments.
Now, that doesn't mean Tildes can't try to address that problem, though. One of my suggestions was "folding" the entire comments section up into its own discrete collapsed (but still expandable) section every X hours/days and/or every Y comments. This mechanism would essentially "reset" the comment sections for important topics on ongoing events, where there is value in consistently providing visibility to new top-level comments since there may be important updates even hours/days later. And I am sure there are other ideas that might work to help the issue as well (like exemplary and noise comment labels help)... but I don't think moving the new top-level comment box is one of them.
I agree but for the reasons you mention this is an intended feature. If a thread gets so long you don't want to scroll down then the chance that your question insight has already been said is significantly higher and your efforts would be much more effectively spent upvoting comments already posted.
If the existing comment tree is so large that reaching the bottom is a hassle, it's unlikely that a new top-level comment is going to be very useful to add. It's generally better for people to get involved in the discussions that are already happening than post a new top-level comment (and that's what the box being at the bottom encourages).