I already posited that LLMs may perform lower-level thinking. Also, in the reference you linked to read their third stated limitation. Perhaps one day some kind of AGI will do “higher-level” aka more realistic thinking… but not yet. Which, I want to add, is not entirely relevant to their utility to us as humans.

Talking about “thinking” gets us distracted from what can actually be done, though if we do want to do it then we’d need to come up with a working definition of what “thinking” actually is.

The difference between the cash register and an inference engine is the training data. Also, barring something like mechanical failure, I would trust the calculation of a cash register (that has previously been demonstrated to function with high fidelity, and barring any major incidents such as dropping it onto the floor that might make me question the relevance of those prior tests to its current performance).

But the results of a LLM “calculation” - and I mean this with total sincerity - might be something like:

“What is 1+1=”?

“Answer: 6, 7!! 🤪”

Or in ye olden times, one expected response might be “ur mother!” or some other flippant remark… exactly like a small child, imitating others might do. Monkey see, monkey do => the principles undergirding LLM technology? It does not understand what it sees, hence it does not “know” when to apply what answer, only going by the most popular (whatever the weighting scheme is - probably highest upvoted answer on Reddit?) string of words that seem to be associated with the question.

He admits himself that nobody really knows how people think, and so all the talk about LLMs looking like they are “thinking” is based purely on vibes rather than reality.

Then again, the thought occurs to me to ponder, how many actual humans really think also? Do 4-year-olds? Do <insert name of opposing political party here>?

That thought experiment helps me to understand. Even if an LLM does perform a kind of “lower thinking” - i.e. mere imitation, of the kind that a monkey or even a 6-year-old human child might - even then, I still would not trust it to vibe-code anything for me.

And it most definitely does not “think think” i.e. higher-level thinking, though again, neither do many humans, seemingly the largest majority of the time.

According to it, the median community replies to 74% of all posts, which doesn’t look bad at all imo.

No you misunderstood - it said “So this is the nine largest communities on each of fourteen instances, taken in subscriber order…”, meaning sorted in descending order of size. In essence then, it found out that even in the absolutely top most active communities across many instances, not all posts will get responded to.

Obviously it will from e.g. !politics@lemmy.world or !memes@lemmy.world, but from more niche communities such as Single-Board Computer Stuff@discuss.online (this is the #9 most active community per month on that instance), which only gets posted to perhaps every other week (and this exclusively from the sole mod, with only one exception that I presume is an alt of the AI bot here) and most of those receive zero or one or two comments, the likelihood of receiving a comment is very low.

Even in our most highly active communities, something like a quarter of posts still get ignored and not responded to. Which means that most of the Threadiverse is super concentrated into only a few communities that are exceedingly active, and then there is everything else that is… um… NOT that.

This is not a sign of health, rather it is a sign of poor community engagement.

If you presume that these posts were worth responding to. Which this bot’s aforementioned post did receive a (single) reply… though only to criticize the bot for having wasted their time by posting slop.

Edit: my guess is that the true medium reply rate would be zero. There are a metric shit-ton of communities that only have 1-2 posts in them period, and none at all from the last month or even year. Like here’s one: https://discuss.online/c/xterra. It supposedly has 61 subscribers, yet has never had a post, ever. The single mod likewise has never once made any posts, not just in that community but anywhere ever, nor commented, nor even voted it would seem (anywhere that hasn’t opted-out of lemvotes.org). It looks like the existence of the community is holding the name just in case someone wants to use it in the future, or even more likely since 3 years ago was the Rexodus, expressing a desire to start up the community, but then that never happened. There are TONS of communities like this, which only exist on paper as the saying goes, rather than in actual reality.

Juicy and delicious 😋

About Tesseract: Admins tend to move exceedingly slowly… but I would guess that at this point given the severity of the breach, most instances might be amenable to at least roll back to an older version before implementation of the hidden shadow list, or perhaps even to switch to using db0’s fork that removes it? Except that db0 is also antagonizing especially LW and possibly even calling for defederation from them at some future date, so I could see strong hesitancy to make stronger tie-ins with him at that point, especially given the sensitivity already surrounding the Admiral Patrick issue - so perhaps they would want to simply deprecate it altogether. It is a quite popular one though.

Most of the discussion that I’ve seen so far is merely bravado e.g. bragging about having made it into the list of shadow-banned users, rather than efforts to move forward to a positive resolution.

So instance admins might welcome such a discussion… perhaps after a cooling-off period first?

However, Tesseract is not a 3rd-party app - it’s an alternative front-end UI for Lemmy. So that means then that i.e. it’s not up to individual users to change the version offered, and instead is solely under control of each individual instance admin that offers that alternative.

Oh that’s so freaking awesome!!! 😎

On the wider scale, I am saddened about the fight between Rimu and you - but I get it, despite how he admitted himself that he did not handle things fully properly (afaik? not that I even begin to comprehend matters involving licensing), I understand if you are just so completely “done” with that whole situation.

I just wanted to say that from my POV I see you two as allies (even though definitely not friends) in the struggle against corporate enshittified platforms.

And I love seeing the development of features that apparently you both agree are quite good - hehe even good enough for Rimu to steal:-P.

Thank you for making your code available as FOSS, and thereby contributing to the growth of the Threadiverse. I’m losing hope that we will ever attract mainstream normies by virtue of continual in-fighting but that has basically nothing to do with you, and in fact PyLova seems actually a force that helps heal the divisions a bit by e.g. decoupling the OG PieFed code from certain authoritarian tendencies lately. Such as shadow-censoring of votes even from Lemmy or Mbin users delivered to PieFed communities, rejected (if I am understanding them correctly) due to the vote quotas that are never once explained to them as a reason for rejection - although tbf Lemmy also offers admins the ability to shadow-censor whole entire communities without mention of that fact either. No software is perfect, but you are continuing to pour your love into this project, which I find awesome and want to say thank you for.:-)

Lemmy most definitely has shadowbans, but they are applied at the level of whole entire communities rather than end-users. I remember being on the StarTrek.Website instance when that functionality was added to the Lemmy code, and discussing with someone my finding out that ChapoTrapHouse was no longer accessible from it, despite the fact the instance remained federated. No announcement of that fact was ever made iirc, it’s just that one day poof the behavior has changed, after Lemmy deployed a version that allowed for that.

Some few instances are very open about their hiding of communities - see e.g. https://legal.programming.dev/docs/hidden-communities/, listing in detail every one - but most are not, and the Lemmy UI has nothing at all like the list that it shows for (de)federations.

The Lemmy admins also can remove people even from the modlogs, in cases of severe abuse. Tbf it’s exceedingly rare and there is little evidence of this being abused (well there was one time, but the excuse that it was due to a bug seemed believable to me and most everyone else - Lemmy.ml quite often tests new versions of the code before they get rolled out to the rest of the Threadiverse, which is quite helpful actually).

Lemmy also fails to notify someone of their having been moderated. The various communities are FULL of stories of people asking where their posts went, or why they can no longer post to a community, or stating their discovery that they have an event in the modlog. The notification of moderation activities is exceedingly poor on the Lemmy platform.

Sadly, Reddit does a much better job at notification of an event, and also provides for a modmail to communicate with the mods to discuss the situation, e.g. to ask why, rather than leaving it to the end-user to wonder if they should DM each one listed in sequence until they find one that is even active much less responsive…

Lemmy is imperfect. At least Huffman is not in charge here, but conversely the tools have aways to go before we catch up even to what Reddit offers.

Given everything that has been going on, would I truly be okay if I were? 😉

REALLY!? Hrm, perhaps I should get curiouser and learn more about this…

Had me in that first half… 😜

It’s dangerous for sure - both collectively (e.g. it has been demonstrated how they can map out military bases and movements following the patterns of when either the or even the surrounding people login, or perhaps do not log in during certain periods of time), and in a sense perhaps individually for people (this is probably most people) who are susceptible to additive behaviors. Although ofc forewarned forearmed and all of that, so just like any source of positive reinforcement (alcohol, drugs, gaming, even unquestionably “healthy” activities such as exercise and working), it needs to be used appropriately i.e. kept in check. “The unexamined life is not worth living”, and all that.

I should note that it was not America that made TikTok that way - at least not directly. That company has always done those activities. Capitalism always sells not what people want but what the company desires to sell them. If we want different - e.g. Loops or like PeetTube and Pixelfed and such - then we have to build it. The code is not simply going to spill out of a random generator and then email itself to us:-) - it requires a staggeringly enormous amount of effort.

And that is the problem. I, like you and most of us here, want an algorithm that isn’t Their Algorithm, but what we end up with instead still ends up being a form of Algorithm, just fueled by upvotes (vulnerable to disinformation) rather than monetary gain.

And we should all disconnect - I would argue not entirely, but stepping back a bit may do us all a world of good. Spend most time with people irl, spend only a little learning here, and don’t neglect other sources that are significantly better as primary or secondary or even tertiary means of getting information. Other than Reddit though, which is where the content creators are, I am not sure more specifically what that may look like but now I know the task ahead of me: to find that.

Georgia… the country? (/s, it’s clearly about the USA state)

It sounds like you might enjoy TikTok. Those Algorithms exist for a reason - they give people what they want. Sometimes they can be overzealous, like you watch just ONE video in YouTube and it starts showing you more like it… but overall people use those platforms because they enjoy what they offer to them.

And as you say, PieFed takes giant leaps towards offering a user power to find what they want, e.g. in providing hashtags, which of course so few people use since even if they submit a post to a community on PieFed that would support them, if you do that from a Lemmy account, or via a 3rd-party app that does not provide (easy) support for hashtags, then the post will have none.

The entire Threadiverse is an experiment. To see what people will build, and to see what the community overall will use. It’s sad to see it struggling so much but… writing code it turns out is so much easier than engendering consensus among people.

In the meantime, I now suggest to people that they not rely upon the Threadiverse and especially Lemmy as their sole source of social media. The tools are still being built, and even then something like 99% of what is already available is simply not being used, and the primary thing is Content - or you might argue rather interactions with people - both of which are fairly lacking here. e.g. so many major events happening irl that I don’t even hear about the tiniest bit about here, and many times I have even searched the Threadiverse for keywords after I’ve learned about the topic elsewhere in case it was here but I simply missed it, only to find that as far as the Threadiverse is concerned the topic simply does not exist. Our low number of content creators is simply insufficient to fully replace Reddit or other sources, unless one’s goal is simply masturbatory dopamine hits from exclusively looking at memes, or arguing about which facts exist or not (were there any fatalities in the Tiannamen Square Massacre or not?).

The Lemmy way: yay!

The second half: yes I understand and agree - it’s a rather complicated issue, which can go both ways. Perhaps it can be something left to the admin to decide whether to enable vs. not? They can see all votes themselves in any case. In general I am not a fan of restricting people’s abilities “for their own good”, which seems in line with the spirit of PyLova, but in any case I definitely understand that it needs thinking about first bc it does run against the grain of what PieFed has done so far.

Thanks for talking about it with me.:-)

What would be more ideal? The tighter focus that e.g. subreddits provided? More niche communities here keep dying off, so that we cannot have e.g. a community based on a single game, so much as a community like just “gaming”.

PieFed has a bunch of features that may help - post flairs (e.g. see posts about the Threadiverse but not Mastodon-style Fediverse posts), user flairs (e.g. see posts by senior “enthusiasts” in a community while excluding all those “hey all I am new and I love the Threadiverse so far”), or do direct keyword filtering. There are also user icons that you can place next to people’s account names to help you visually spot posts from people that you might want to either specifically read or perhaps avoid, but whom you did not want to simply block. You can also “follow” users too, receiving notifications whenever they post.

Also, since you can always find content in those categories of communities / Topic Feeds, you can fine-tune your actual subscriptions more so that you make it tighter in the general sense. i.e. I’m saying that you can both have your cake and eat it too! e.g. I unsubscribed from the most popular politics communities, which keeps it out of my Subscribed feed and allows other posts - even low-interaction stuff (with <10 upvotes) like poetry - to come through, but then whenever I want more I can always view those posts from a politics Feed, so nothing is “lost” that way.

PieFed offers a significantly more robust set of tools than Lemmy. Though in general the entire Threadiverse is niche, and cannot replace what Reddit had - like better post organization (across communities) is great and all, but when the actual content creators choose to submit to Reddit but not here, then there is only so much that can be done about that.

The Threadiverse isn’t like “dying” or anything, but it’s definitely not thriving either, when we used to have 65k active users while now we have 35k - people are leaving us in droves and not coming back.

Yes that - the second one, as shown in your image - is what I mean. Lemmy offers the option to show upvotes separately from (& adjacent to) downvotes. PieFed’s tooltip approach fails to work for me something like a fifth of the time, at least on PieFed.social. At least a third of the time it will not work even if I click the number >10 times at varying cadence, although most of the time then if I zoom in and carefully hit precisely the number rather than the general area it is in it will work. Ofc that could be coincidence like a server connection issue that manages to work upon subsequent clicks regardless of how I approached hitting the number.

Regarding showing who delivers the downvotes - that can be quite useful sometimes as well. I just had a serial downvoter go through every comment that I had made recently, across a wide swath of communities from memes to Fediverse matters and news and politics etc. - ironically they had never once posted and only made 3 comments in I think it was 3 years, yet they had downvoted people sth like 20k times and only upvoted iirc <3k total, so I seriously doubt that I was the only one that happened to. I want negative feedback the same as I want positive ones, but like… delivered as an adult, using words (at which point others piling them on only adds assent & therefore weight to that spoken position), not this childish “I don’t like your politics so I am going to downvote everything you’ve said for days, without even telling you why”.

Voting really is an imbalance of power if someone can do things like that, but the recipients have no power to block it. It has a chilling effect upon social media bc people will either speak less often, knowing they will receive those, or else grow such a thick skin that even a veritable flood of downvotes affects them not in the slightest. I don’t like either of those options.

Though I suppose there is an argument that people could too casually block others for offering downvotes… or something? Honestly I don’t quite understand that - anyone can find out who it is, with just a little bit of effort (e.g. I could make a post in a community that I moderate, so then when my serial downvoter strikes then as a mod I can see who they are, even if not as a user; or ultimately if I installed my own server then I could casually inspect the votes of everyone on every post or comment everywhere), this information is not “private” and there seems little reason to not place tools into people’s hands to balance out this otherwise inequitable relationship?

But anyway I’m not pushing for it - I just hate to see how like powerful people (e.g. jordanlund) have all this ability to see and do things (e.g. see people who downvote and block them), but what can the common person do? Although some of this seems to be based on the misunderstanding that accounts that you have specifically blocked cannot still vote on your content, whereas now I think that they can. There seems nothing that can be done about that.

Btw Mbin shows who offers (links given for a randomly chosen example post): upvotes aka favorites, separately also Increases, now seemingly called Boosts, but will only show the total count but not who offered them for Reduces (I don’t know if a mod can see these, obviously an admin could but I don’t know whether via this same interface).

Impressive!

I did get excited thinking that it might show downvotes, but now I see that’s only for admins.

Do you plan to expose downvotes received on each post/comment to the user? I’ve been through every single theme and setting on PieFed.social and could never find a way to display those - although I don’t know whether other PieFed instances can already do differently or not (they definitely don’t show by default in places like PieFed.zip).

Fwiw I think people would be very happy to see their downvotes separately from upvotes, since e.g. +10/-9 is quite a bit different scenario than simply =1 overall (controversial vs. total lack of engagement). Technically this is already possible but the tooltip display is exceedingly janky and does not work at all a good fraction of the time, when when zoomed in like 10x (which helps at least sometimes).

Anyway it’s always neat to see a “new project”, even if as a fork this is only kind-of half of one, it still counts in terms of offering new opportunities to users, which is always a good thing.

Or alternatively: It shows that handle on one specified instance, e.g. always from your Mastodon home instance.

Fwiw, PieFed already translates URLs on the backend so that visiting content from a foreign source will take you instead to the local copy.

Also this post might be tangentially interesting: https://piefed.social/post/1285215.

Edit: oh derp, somehow I entirely misread your message as that you had not heard of Friendica’s entire existence. Sorry, ignore this if you like, or perhaps it’s interesting in a different manner now. 😜

Your instance is quite a bit cut off from the rest of the Fediverse, although even so the news of Friendica has made waves - e.g. https://beehaw.org/post/521802 with >800 comments.

Friendica is fairly niche though, and suffers from a lot of problems. For one, as a Facebook replacement it relies upon your family members and close friend group to also use it - but unless you act as tech support and install it onto their machines for them (alongside Linux and everything needed for them to self-host), then how else will you be able to communicate with them on that platform?

And it more or less does need to be either self-hosted or at least hosted exceedingly often, since it does not scale well and users of the project have started begging for technical support from outsiders (not a great example of the latter but see this informational post).

In short, there are reasons for its lack of spread so far. I do not know if those issues are solvable - though by now perhaps they already have been? It’s just a very different model from the more semi-anonymous platforms like the Fediverse, which can make things challenging.