JustTesting

@JustTesting@lemmy.hogru.ch

• ⁨Comment⁩ on ⁨Why do they turn Federation into a dystopia?⁩ ⁨⁨2⁩ ⁨days⁩ ago⁩:

  I think there’s still a lot of room to explore without abandoning the utopia setting. like we usually only see the spaceship stuff, but what about a more political drama taking place on member worlds, that kind of thing, i think it could be amazing.

  also, as you say, it’s been done for 60 years. Might as well do the same thing over again for a new generation that hasn’t seen tos/tng/ds9. They don’t know it yet, so it’s not overused, and the TOS audience wouldn’t be the target audience anyways. and could still explore new topics. the audience isn’t the same, our world isn’t the same, making the same show again would still not be boring as it be a completely different thing.

  Both approaches can work imo and have a place, without the need to go more dystopia.

• ⁨Comment⁩ on ⁨Ad blocking is alive and well, despite Chrome's attempts to make it harder⁩ ⁨⁨5⁩ ⁨days⁩ ago⁩:

  Plus firefox mostly exists because google pays them, probably so there’s no anti-trust action against them.

• ⁨Comment⁩ on ⁨A lot of the laid-off staff from the Washington Post should start a news cooperative. Seriously!⁩ ⁨⁨6⁩ ⁨days⁩ ago⁩:

  Don’t know about the second point, but on the first, there’s a online newspaper here that does it pretty well. It’s like 240$ a year, but with the option to pay however little/much you want. the articles can be shared freely (no paywall, though i think since a year ago you need to enter your email to read, used to be completely free to share), but can’t be discovered/found unless you’re subscibed. it’s split into two legal entities, the newspaper that employs the journalists and a second non-profit that actually collects the payments and that every subscriber is allowed to vote in, elect leadership for etc. that works out guidelines for the newspaper part to follow.

  has been working pretty well for several years now and it’s one of the last few places of quality, independent journalism in my country

• ⁨Comment⁩ on ⁨AI agents now have their own Reddit-style social network, and it's getting weird fast⁩ ⁨⁨1⁩ ⁨week⁩ ago⁩:

  Ah but don’t worry, there’s also skills for scanning skills for security risks, so all good /s

• ⁨Comment⁩ on ⁨AI agents now have their own Reddit-style social network, and it's getting weird fast⁩ ⁨⁨1⁩ ⁨week⁩ ago⁩:

  They also have a ‘skill’ sharing page (a skill is just a text document with instructions) and depending on config, the bot can search for and ‘install’ new skills on its own. and agyone can upload a skill. So supply chain attacks are an option, too.

• ⁨Comment⁩ on ⁨Pope Leo XIV brings not peace but a sword to AI oligarchs and a slop-mad world in new address, says it's 'Turning people into passive consumers of unthought thoughts'⁩ ⁨⁨1⁩ ⁨week⁩ ago⁩:

  I got mine from kultofathena.com 20 years ago for 40 bucks (though with fancier blade), but just checked and don’t think they sell it anymore.

  Image

• ⁨Comment⁩ on ⁨Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task – MIT Media Lab⁩ ⁨⁨2⁩ ⁨weeks⁩ ago⁩:

  I’m sorry, that came off very passive-aggressive, I really shouldn’t post at 3 am when I can sleep.

  The whole dark ages, and golden ages thing is just very annoying, made up during the renaissance, in part as a useful tool to go “look how shit everything is, I will make it great and amazing like it was before”, still a favorite to use by populists (in reference to whatever time is most suitable) and it’s been repeated so much, it actually works, everyone kinda just accepted it. But then when you dig into it, the middle ages weren’t really worse in terms of invention/art/etc. than the renaissance, nor was there this big stagnation after the decline of the roman empire, and people always made art, new inventions and great achievements, along with cruelty, bloodshed and other awful things. But then this has been a relatively recent shift in historical research, so not that well known

• ⁨Comment⁩ on ⁨Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task – MIT Media Lab⁩ ⁨⁨2⁩ ⁨weeks⁩ ago⁩:

  In other news, dark ages are a myth disproven by science.

• ⁨Comment⁩ on ⁨What steps can be taken to prevent AI training and scraping of my public facing website?⁩ ⁨⁨1⁩ ⁨month⁩ ago⁩:

  A big issue is that this works for bots that announce themselves as such, but there’s lots that pretend to be regular users, with fake user agents and ips selected from a random pool with each ip only sending like 1-3 request/day, but overall many thousands of requests. In my experience a lot of them are from huawei and tencent cloud/ASN

• ⁨Comment⁩ on ⁨Why ActivityPub over Nostr? - function only⁩ ⁨⁨1⁩ ⁨month⁩ ago⁩:

  I don’t actually know how nostr deals with messages if you’re offline, if at all, not that familiar with the protocol. But your idea sounds workable.

  I tend to come at it from the other side, I like the federated model, but think the “supernodes” could behave more like dedicated relays. Like, a lemmy server right now does a lot of things, like serve a frontend, do expensive database queries to show a sorted feed, etc. and a lot of that does not scale very well. So having different kinds of nodes with more specialization, while still following a federated model makes sense to me. Right now if one of my users subscribes to some community, that community’s instance will start spamming my instance with updates nonstop, even though that user might not be active or might not even read that community anymore. It would be nicer if there was some kind of beefy instance I could request this data from if necessary, without getting each and every update even though 90% of it might never be viewed. But keeping individual instances that could have their own community and themes, or just be hosted for you and your friends to reduce the burden on non-techies having to self-host something.

  Or put another way, instead of making the relays more instance-y, embrace the super instances and make them more relay-y, but tailor made for that job and still hostable by anyone, if they want to spend on the hardware. But I’m still not clear on where you’d draw the line/how exactly you’d split the responsibility. For lemmy, instead of sending 100’s of requests in parallel for each thing that happens, a super-instance could just consolidate all the events and send them as single big requests/batches to sub-instances and maybe that’s a good place to draw the line?

• ⁨Comment⁩ on ⁨Guarding My Git Forge Against AI Scrapers⁩ ⁨⁨1⁩ ⁨month⁩ ago⁩:

  it’s [iocaine)(docs.rs/crate/iocaine/latest) not Locaine, tripped me up at first as well.

• ⁨Comment⁩ on ⁨Why ActivityPub over Nostr? - function only⁩ ⁨⁨1⁩ ⁨month⁩ ago⁩:

  this article and the accompanying discussion from lobsters is very relevant. Though the article itself is a bit one sided in favor of nostr, it doesn’t do a great job arguing why a relay really is better

• ⁨Comment⁩ on ⁨I Went All-In on AI. The MIT Study Is Right.⁩ ⁨⁨2⁩ ⁨months⁩ ago⁩:

  so the obvious solution is to just have humans execute our code manually. Grab a pen and some crayons, go through it step by step and write variable values on the paper and draw the interface with the crayons and show it on a webcam or something. And they can fill in the gaps with what they think the code in question is supposed to do. easy!

• ⁨Comment⁩ on ⁨Anubis is awesome and I want to talk aout it⁩ ⁨⁨2⁩ ⁨months⁩ ago⁩:

  You mean for the referer part? Of course you don’t want it for all urls and there’s some legitimate cases. I have that on specific urls where it’s highly unlikely, not every url. E.g. a direct link to a single comment in lemmy, plus whitelisting logged-in users. Plus a limit, like >3 times an hour before a ban.

  It’s a pretty consistent bot pattern, they will go to some subsubpage with no referer with no prior traffic from that it, and then no other traffic from that ip after that for a bit (since they cycle though ip’s on each request) but you will get a ton of these requests across all ips they use. It was one of the most common patterns i saw when i followed the logs for a while.

• ⁨Comment⁩ on ⁨Anubis is awesome and I want to talk aout it⁩ ⁨⁨2⁩ ⁨months⁩ ago⁩:

  This is the way. I also have rules for hits to url, without a referer, that should never be hit without a referer, with some threshold to account for a user hitting F5. Plus a whitelist of real users (ones that got a 200 on a login endpoint).

  then there’s ratelimiting and banning ip’s that hit the ratelimit regularly.

  Dowloading abuse ip lists nightly and banning those, that’s around 60k abusive ip’s gone. At that point you probably need to use nftables though, for the sets, as having 60k rules would be a bad idea.

  there’s lists of all datacenter ip ranges out there, so you could block as well, though that’s a pretty nuclear option, so better make sure traffic you want is whitelisted. E.g. for lemmy, you can get a list of the ips of all other instances nightly, so you don’t accidentally block them. Lemmy traffic is very spammy…

  there’s so much that can be done with f2b and a bit of scripting/writing filters

• ⁨Comment⁩ on ⁨⁩ ⁨⁨3⁩ ⁨months⁩ ago⁩:

  DS9 has lots of stuff against racism, either with aliens or black people in time travel episodes. And the one where Quark transitions for an episode and it’s not just milked for laughs or similar, for the 90s that was handled pretty tactfully

• ⁨Comment⁩ on ⁨Does anyone else notice an up tick in hostility on Lemmy lately?⁩ ⁨⁨3⁩ ⁨months⁩ ago⁩:

  I’d be really curios to see some sort of study done on this. I mean, it’s not just americans and most of the west is not insulated from america, either, at least not online. and you don’t know from talking to someone online where they’re from. At the same time, there’s rising fascism and neoliberalism bullshit in europe, too.

  I’d love to know how much of it is people getting antsier in general because they’re in a shit situation and how much it’s ‘infectious’ from talking with people in shit situations elsewhere, spreading bad vibes. Is this also happening in the chinese web? How about other countries that are more politically/economically aligned with the west but culturally less part of the english speaking web?

  There has to be some sociologist out there somewhere studying this, no? But i wouldn’t know where to look. if anyone knows of something along those lines, i’d love to hear it.

• ⁨Comment⁩ on ⁨New Study: Global Fertility Rate Decline Now Linked Directly to the Commodification of Housing⁩ ⁨⁨3⁩ ⁨months⁩ ago⁩:

  Where is the new study? They have sources at the bottom but which one is the actual study they talk about?

• ⁨Comment⁩ on ⁨A.I. Video Generators Are Now So Good You Can No Longer Trust Your Eyes⁩ ⁨⁨3⁩ ⁨months⁩ ago⁩:

  But what if your phone comes with nice AI filters? The fake videos get more and more real and the real videos get more and more fake

• ⁨Comment⁩ on ⁨Is Star Trek Discovery that bad?⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  It’d be more accurately titled Star Trek: Burnham

  I always called it ‘The Burnham Show, starring Michael Burnham’

• ⁨Comment⁩ on ⁨⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  I don’t disagree. I meant for users it is incidental. Most users probably wouldn’t buy them with spying as the main purpose(they just also don’t really care that it can spy). making them much more widespread than something where spying was the main use-case, making the problem worse.

  And for Meta it’s like tracking cookies on crack

• ⁨Comment⁩ on ⁨⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  sure, but there the spying is the purpose, whereas with the glasses it’s incidental.

  you don’t buy such gadgets if you don’t intend to spy, but people would buy meta glasses for other reason, and meta being able to spy on you is just a side-effect. Plus it’ a matter of scale, this has the potential of being much more prominent than some spy camera.

• ⁨Comment⁩ on ⁨Get ready to see ads on your… Samsung refrigerator⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  I remember reading that hotel TVs are an option. They also have an ad platform, but one intended for the hotel owner to send ads from, not some 3rd party. Not exactly dumb but also not as bad as regular TVs.

  And of course a beamer or PC screen connected to some cheap small form factor PC is always an option, with Kodi or similar on it, i haven’t owned a TV in like 10 years, just using a small linux pc with beamer, and a tv tuner card in the past (nowadays my ISP offers all public channels on IPTV)

• ⁨Comment⁩ on ⁨⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  For the byte pair encoding (how those tokens get created) i think bpemb.h-its.org does a good job at giving an overview. after that i’d say self attention from 2017 is the seminal work that all of this is based on, and the most crucial to understand. jtlicardo.com/blog/self-attention-mechanism does a good job of explaining it. And jalammar.github.io/illustrated-transformer/ is probably the best explanation of a transformer architecture (llms) out there. Transformers are made up of a lot of self attention.

  it does help if you know how matrix multiplications work, and how the backpropagation algorithm is used to train these things. i don’t know of a good easy explanation off the top of my head but xnought.github.io/backprop-explainer/ looks quite good.

  and that’s kinda it, you just make the transformers bigger, with more weight, pluck on a lot of engineering around them, like being able to run code and making it run more efficientls, exploit thousands of poor workers to fine tune it better with human feedback, and repeat that every 6-12 month for ever so it can stay up to date.

• ⁨Comment⁩ on ⁨⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  Well each token has a vector. So ‘co’ might be [0.8,0.3,0.7] just instead of 3 numbers it’s like 100-1000 long. And each token has a different such vector. Initially, those are just randomly generated. But the training algorithm is allowed to slowly modify them during training, pulling them this way and that, whichever way yields better results during training. So while for us, ‘th’ and ‘the’ are obviously related, for a model no such relation is given. It just sees random vectors and the training reorganizes them tho slowly have some structure. So who’s to say if for the model ‘d’, ‘da’ and ‘co’ are in the same general area (similar vectors) whereas ‘de’ could be in the opposite direction. Here’s an example of what this actually looks like. Tokens can be quite long, depending how common they are, here it’s ones related to disease-y terms ending up close together, as similar things tend to cluster at this step. You might have an place where it’s just common town name suffixes clustered close to each other.

  and all of this is just what gets input into the llm, essentially a preprocessing step. So imagine someone gave you a picture like the above, but instead of each dot having some label, it just had a unique color. And then they give you lists of different colored dots and ask you what color the next dot should be. You need to figure out the rules yourself, come up with more and more intricate rules that are correct the most. That’s kinda what an LLM does. To it, ‘da’ and ‘de’ could be identical dots in the same location or completely differents

  plus of course that’s before the llm not actually knowing what a letter or a word or counting is. But it does know that 5.6.1.5.4.3 is most likely followed by 7.7.2.9.7(simplilied representation), which when translating back, that maps to ‘there are 3 r’s in strawberry’. it’s actually quite amazing that they can get it halfway right given how they work, just based on ‘learning’ how text structure works.

  but so in this example, us state-y tokens are probably close together, ‘d’ is somewhere else, the relation between ‘d’ and different state-y tokens is not at all clear, plus other tokens making up the full state names could be who knows where. And tien there’s whatever the model does on top of that with the data.

  for a human it’s easy, just split by letters and count. For an llm it’s trying to correlate lots of different and somewhat unrelated things to their ‘d-ness’, so to speak

• ⁨Comment⁩ on ⁨⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  Huh that actually does sound like a good use-case of LLMs. Making it easier to weed out cheaters.

• ⁨Comment⁩ on ⁨⁩ ⁨⁨4⁩ ⁨months⁩ ago⁩:

  They don’t look at it letter by letter but in tokens, which are automatically generated separately based on occurrence. So while ‘z’ could be it’s own token, ‘ne’ or even ‘the’ could be treated as a single token vector. of course, ‘e’ would still be a separate token when it occurs in isolation. You could even have ‘le’ and ‘let’ as separate tokens, afaik. And each token is just a vector of numbers, like 300 or 1000 numbers that represent that token in a vector space. So ‘de’ and ‘e’ could be completely different and dissimilar vectors.

  so ‘delaware’ could look to an llm more like de-la-w-are or similar.

  of course you could train it to figure out letter counts based on those tokens with a lot of training data, though that could lower performance on other tasks and counting letters just isn’t that important, i guess, compared to other stuff

• ⁨Comment⁩ on ⁨'3d-printing a screw' is a way to describe how AI integration is stupid most of the time⁩ ⁨⁨5⁩ ⁨months⁩ ago⁩:

  one other use case where they’re helpful is ‘translation’. Like i have a docker compose file and want a helm chart/kubernetes yaml files for the same thing. It can get you like 80% there, and save you a lot of yaml typing.

  Wont work well if it’s mo than like 5 services or if you wanted to translate a whole code base from one language to another. But converting one kind of file to another one with a different language or technology can work ok. Anything to write less yaml…

• ⁨Comment⁩ on ⁨Wealth inequality seems like the only outcome in a system where capital gains are taxed less than labor⁩ ⁨⁨5⁩ ⁨months⁩ ago⁩:

  Wouldn’t that just lead to splitting off of cheap companies, with pro-bono ceos that get paid more by the parent company through side channels? I don’t think there’s any fixing it with these kinds of laws, as they’ll just find loop holes to circumvent it.

  maybe if companies were forced to be democratic so figurehead ceos could be ousted by the underpaid workers, but at that point it’s not capitalism, but socialism. and that’s how it usually goes imo, the workable solution to capitalism turns out to be not-capitalism

• ⁨Comment⁩ on ⁨Sam Altman admits OpenAI ‘totally screwed up’ its GPT-5 launch and says the company will spend trillions of dollars on data centers⁩ ⁨⁨5⁩ ⁨months⁩ ago⁩:

  I’m not really sure I follow.

  Just to be clear, I’m not justifying anything, and I’m not involved in those projects. But the examples I know concern LLMs customized/fine-tuned for clients for specific projects (so not used by others), and those clients asking to have confidence scores, people on our side saying that it’s possible but that it wouldn’t actually say anything about actual confidence/certainty, since the models don’t have any confidence metric beyond “how likely is the next token given these previous tokens” and the clients going “that’s fine, we want it anyways”.

  And if you ask me, LLMs shouldn’t be used for any of the stuff it’s used for there. It just cracks me up when the solution to “the lying machine is lying to me” is to ask the lying machine how much it’s lying. And when you tell them “it’ll lie about that too” they go “yeah, ok, that’s fine”.

  And making shit up is the whole functionality of LLMs, there’s nothing there other than that. It just can make shit up pretty well sometimes.