Gyokuyoutama

Oh shit, it accepts JP inputs despite the claims.

However, it still bans terms.  Even 尻尾 (tail) is banned, probably because the first character means "butt."

You get different results but they aren't exactly more reliable.  For example, here's what I get for 八雲藍, 東方 (Ran Yakumo, Touhou)

Well, it's got the Touhou vibe down, but there's no sign of Ran particularly.  Let's try:

八雲藍, 東方, 狐, 天狐, 式神, 九尾 (Ran Yakumo, Touhou, fox, tenko, shikigami, kyuubi)

It's definitely went the kyuubi route, but we've lost all sign of Touhou project and this still isn't Ran.  Maybe it needs a description, so let's try:

八雲藍は橙と散歩している, 東方 (Ran Yakumo is on a walk with Chen, Touhou)

Well we're in the Touhou wheelhouse again and we do have the correct scene, but these still aren't recognizable characters and certainly not the Yakumos.  Let's try something more mainstream (and I also mixed up the grammar 

霊夢と話している魔理沙, 東方 (Marisa talking with Reimu, Touhou)

Okay so even in Japanese it knows Touhou style but doesn't know the characters.  They are talking though so it is parsing the Japanese grammar.  Let's try something more mainstream:

空条承太郎, ジョジョの奇妙な冒険 (Jotaro Kujo, Jojo's Bizarre Adventure)

So yeah, this is definitely a "doesn't know Touhou characters" problem not a "can't parse Japanese text" problem.

Japanese searches seem like they dig into more untapped regions of the data set.  For example here's what you get for "Tomoko Kuroki, Watamote" in English:

It's clearly pulling the title of the manga (though not quite correctly), but otherwise this isn't very Watamote related.  Here's a result from the equivalent Japanese search, 黒木智子, ワタモテ

We have the title appearing again, but now it's a black and white manga with a character who still isn't Tomoko, but at least getting into the same ballpark.  There were other results for the Japanese one and they were all black and white manga of a nerdy girl in a cluttered room.  (They all had glasses for some reason.)  Some had even more mangled Japanese text which might have been part of the title.  In contrast the English search is all cutesy color images.  The same thing happened with Jotaro by the way: Search in Japnase, get black and white, search in English, get color.  Except there you get Jotaro either way.

I don't know what to do with this knowledge.  There's still a large number of forbidden words even in Japanese, the output frequently gets censored as inappropriate, and I haven't been able to find anything that it can successfully draw in Japanese that it couldn't draw in English.  The only real difference seems to be that if something is based on a manga, the JP searches are derived from the manga artwork while the English searches are derived from the anime artwork.

EDIT: For example, here's what I get when I put in every single variation of Girl und Panzer's name and every way of referring to Erwin (including her real name  松本里子 - Satoko Matsumoto)

That is, the same generic anime girl in front of a tank that I got when I did similar prompts in English.

Pour one out for RightStuff anime.

As far as I can tell Discotek only distributed their stuff through RightStuf and Amazon, so I guess I'm back to Amazon.

On 10/8/2023 at 12:35 PM, Moby said:

Was chatting with my parents and brothers during coffee, when the conversation topic got to "little kids with unsupervised access to phones and PCs" because of my neighbor's loud and obnoxious kids.

 

Me and my brothers only got our first phone at 15 (my parents got their first ones at almost 40), barely used the internet back then due the lack of technology and we are arguably normal people, while they were given PCs and phones since early age and got diagnosed with autism. Then I remembered, my other young cousins that also were given phones since early age and had little parental care also were diagnosed with autism.

 

That got me thinking, does the combination of mind rot and overstimulation from constant unsupervised internet access, plus lack of parental care, causes autism?

 

Now, I am not saying "phone radiation causes autism", but just wondering that if someone spread that "unsupervised phone and internet access to kids causes autism", we would have less kids being raised with phone glued to their face.

 

....Nah, nothing would change.

I've been dealing with students for about a decade now and I can say that the number of students with obvious signs of autism have been steadily increasing.  In particular:

• Avoidance of eye contact at all costs.
• Use of set phrases in all interactions while obvious not meaning them.  For example I had a student who would always begin conversations with me by mumbling "Hello, how is your day going?" and when I said the obvious "fine" or whatever he would always mumble "okay, I'm glad" before getting into what he really wanted to talk about.  This set pattern would happen even if I initiated by saying something like "so, what's on your mind today?"
• Very obvious irritation when being asked to wait, ex. if I have four students asking me questions and I talk to someone else first.  Things like pacing frantically or even banging on the wall.
• The good old swaying in the seat while talking.

That being said this is still maybe like 5% of the students I deal with.  It's just that when I started it would be very rare to deal with someone I thought was autistic, and now it's just "oh, it's another case of autism."  On the other hand the usage of phones has absolutely exploded.  Pretty much every student is using one all the time, unless they are involved in an immediate conversation.  So phones definitely aren't a sufficient cause for autism, but I wouldn't be surprised if they added to the prevalence of it.

What I have noticed on a universal scale is drops in attention spans and increased anxiety when dealing with confusion.  The attitude is "if I don't know it immediately I must find it online, or else I'm screwed."

I'm getting the "unsafe content" warning significantly less often than I was a couple hours ago and coincidentally the load on the system is much less than it was a couple hours ago.

My best guess is this (and it is just a guess): the "is it porn?" bot is probably CPU intensive and has a timeout feature of "yeah, if it's taking you this long to figure it out, it's probably porn."

Anyway, as image tax, here's some monster girls in goofy situations:

 

I also tried to do the "Harpy can't open a bottle with wing nubs" but I couldn't get it to output a harpy with wings for arms.

 

7 hours ago, A 1970 Corvette said:

The only possible explanation is that Suburban Sasquatch is simply that influential of a movie that the search algorithm prefers it. Or the algorithm's fledgling sentience is first becoming apparent by having very picky taste in movies and vtubers

Really I have no idea what was going on, since I found out late in my investigation that a lot of the behavior was inconsistent.  For example, "Suburban Sasquatch" now only returns videos actually from the channel.  Most VTuber names will usually not return extra results, but then randomly work.  It's like the search just randomly decides that it wants to restrict itself to the channel or not every time you search.

However, results for movies like Terminator: Dark Fate or most Star Wars movies (but not Attack of the Clones or Revenge of the Sith for some reason) will still reliably return the top results.  My suspicion is that big companies paid for "targeted searches" or whatever, like they do on Amazon, so those ones were actually paid for.  And the implementation of the system somehow broke searches for terms not on the "targeted advertising" list.

Weirdest note about checking today: When I searched "Eilene" yesterday it would not only reliably give me results off the channel, but they'd actually be about the Vtuber.  In retrospect that's doubly weird because that's not what you get on a fresh browser's search for "Eilene;" instead you get "Come on Eileen" and the artist Eleine, despite neither being spelled the same way.  Today when I search for "Eilene" in channel I get those results instead.

Outside of the results for the big movies, which are reliable, I really do think this is just from Youtube search acting in ways that no one, not even Google, understands and generally being broken.

My guess from playing around with things is some combination of the following:

• The "we detected the output is unsafe" and the "we have too many requests at this time" messages seem to show up for me in about the same contexts with the same types of prompts.  This leads me to suspect some bullshit, either that some of the "we're too busy" messages are really due to the content, or some of the content messages are really there just to save CPU in peak times, or some mixture of both.
• I've ran prompts that have gotten blocked 10 times and then produced four images on the eleventh attempt, so I doubt that the deciding factor is the input.  (I mean, the input is a factor in that certain terms are blocked, sometimes dependent on the context, but you get a different message for that before it starts generating anything.)
• I suspect that there is a separate AI that runs a "inappropriate content" check on the output, kind of like those "kid safe browser" tools.  Just like those there's probably a lot of false positives and occasionally some things that really should have been blocked but weren't.
• I assume that any time you get less than four images it's because it triggered the post creation filter.  I pretty much never get four images, even when both the prompt and the images that actually did output are really tame, so I assume that the post generation check is weighted towards blocking things.
• That being said, in the four or so days I've been using it I have seen a much higher fail rate in the last couple of days.  Previously I could swap to an entirely new prompt without anime refences (or other things were the training data might have been lewd) and get several results on the first go.  Now damned near everything will run into the "unsafe content" warning at least every other attempt.  So my guess is that they've just been steadily ramping up the sensitivity on the post creation censor bot.
• There are apparently things that you can do that get you flat out temp-banned, such as making pictures of Bill Gates or using Pippa's @ workaround.  I've also heard that making "historical" pictures will lead you into trouble.  I haven't had that stuff happen to me personally though.  But this seems to be at third system independent of both the post creation censor bot and the banned term bot.

But since this is theoertically an image thread, have the closest thing to Tomoko I got it to output:

Further investigations.  Vtuber search terms: everything I tried won't bring up videos from off the channel, except:

Roboco, Inugami Korone, Usada Pekora, Shiranui Flare, Shirogane Noel, Uruha Rushia, Himemori Luna, Mano Aloe, Laplus Darkness, Takane Lui, Sakamata Chloe, Kazama Iroha, Ayunda Risu, Airani Iofifteen, Kureji Ollie, Anya Melfissa, Vestia Zeta, Kaela Kovalskia, Ceres Fauna, Nanashi Mumei, Hakos Baelz, Tsukumo Sana, Shiori Novella, Kouseki Bijou, Nerissa Ravencroft, Noir Vesper, Yagoo

Phase Connect, Tenma Maemi, Fujikura Uruka, Kaneko Lumi, Himemiya Rie, Komachi Panko

Projekt Melody, Kson, Amemiya Nazuna, Henya the Genius, GEEGA, Nyatasha Nyanners, Apricot (the results aren't the vtuber though, but it's still notable that any results come up rather than a "no videos" result)

Nijisanji (presumably some of their talent would also be hits, but I ain't searching all those.), Eilene, Natsumi Moe, Raven Manor

This may seem like a lot, but I tried every single name in Hololive (as well as "Hololive" itself), Phase Connect, VShojo and every classic big name Vtuber I could think of, as well as some very obscure ones (ex. Nanahoshi Suzu.)  So this actually represents a small minority of the search terms attempted; most will not lead to results off the channel.

By experimenting on searching on Pippa's channel I can see that some of these are "category one" and some are "category two" in that some pull up videos on her channel and some do not.  For example "inugami korone" brings up a video about a dog on her channel while "Usada Pekora" doesn't bring up her reading of the "Let's pretend I'm still Pekora" greentext.

I chose Vtubers because I thought maybe the movie stuff was getting dominated by youtube finding movie names in some database, and maybe that database just had things like "Star Wars" and "Attack of the Clones" omitted by chance.  The chances that Vtubers would get special privileges is low.  But there was still variation in how the searches went through.

And actually now I realized that I might have wasted my time classifying this, since when I tested some names on the list a second time, the behavior changed.  For example, "Inugami Korone" doesn't get off channel videos for me now despite doing that on several channels when I started doing these searches fifteen minutes ago.  "Fujikura Uruka" no longer works, but "Tenma Maemi" does, despite both being about the same popularity from the same company and both working at the start of the searching.

So it looks like at this level the search algorithm might just be breaking somewhat randomly.  But it's not completely random, since many of the above names have been consistent and the names that I'm trying from off that list still don't work.

TL;DR: Youtube search is broken in a way which cannot be categorized properly, and will lead you to madness if you attempt to do so.

For some reason searching for certain movie titles in a youtube channel search will not pull up videos from that channel but instead do a general search of the site.  I discovered this when trying to see if RLM did a Dark Fate review.  They did, but if you search "Dark Fate" on their channels you get a bunch of videos from Paramount Pictures and various movie clip channels.  Note that searching "Terminator" does get you videos just from the channel (with the review in question coming up second) though "Terminator: Dark Fate" gives you a general search.

This seems to be consistent across every channel I tried it on.  "Dark Fate" gives you the same results regardless of the channel.  Now what's interesting is that if you search "The Terminator" on channels with terminator videos, those come up.  But if you search "The Terminator" on channels with no terminator videos, then you get links to watch the movies on Youtube as well as some various other terminator videos from random channels.  In contrast if you search for some nonsense like "Cyborg Bigfoot" you will get the message "this channel has no content that matched 'Cyborg Bigfoot'" despite there actually being content on youtube as a whole for that topic.

After some investigating I found basically three categories of movie titles: 1.) those that would always turn up results unrelated to the channel, even if the channel had related content, 2.) those that would turn up some videos from the channel and lots not from it (you can check that the channel matters by doing the same search on different channels with and without relevant content) and 3.) those that would only return results from the channel, returning nothing if there was no related content.

Category 1 (Always returns videos not from the channel) - Dark Fate, Terminator Dark Fate, Rise of the Machines, Genisys, Judgement Day (though it returned WWE results), A New Hope, The Empires Strikes Back, Return of the Jedi, The Phantom Menace, The Force Awakens, Raiders of the Lost Ark, Temple of Doom, The Last Crusade, The Dial of Destiny, Jurassic Park, The Shawshank Redemption, Suburban Sasquatch

Category 2 (Returns videos from the channel and stuff not on the channel) - The Terminator, Terminator (removing "the" resulted in fewer results from off the channel), The Last Jedi, The Rise of Skywalker, Indiana Jones, Kingdom of the Crystal Skull, Back to the Future

Category 3 (Never returned results from off the channel) -  Star Wars, Attack of the Clones, Revenge of the Sith, The Last Vampire on Earth, Tammy and the T-Rex

I have no idea what the hell this means.  The fact that several Star Wars movies aren't given the special search treatment but Suburban Sasquatch is is particularly baffling.

I really wonder how the "inapporpriate output" filter works.  Sometimes it repeatedly blocks prompts like "Character X orders food at McDonalds" due to the output being inapprporiate.  And then later it outputs stuff like this:

Some weird phenomena I noted:

It knows what Heavy Weapons Guy is but not what his face looks like.  Also, if you put Rainbow Dash and Hatsune Miku in the same picture they often end up fused together.  Here is a picture where both things happened:

Supposed to be a tea party with Rainbow Dash, Hatsune Miku, Sailor Jupiter and Heavy Weapons Guy.  Somehow we got Mini moon, two copies of a fusion between Hatsune Miku and Rainbow Dash, a heavy weapons guy cosplayer, and some random dude.

Other variations of that attempt showed me that Rainbow Dash very often comes in pairs, and that it knows Touhou fashion but not Touhou characters:

This time Heavy is a steampunk demon.  Twilight is supposed to be Marisa from Touhou, which she doesn't match at all, but that hat definitely could fit in a ZUN game.  Mob hats are also very common when asking for Touhou, even though the characters never match up.

It definitely knows MLP by itself though.  For one example, here's Twilight as an angry forum moderator:

It reliably made Disney's Maid Marian a fox, but always put her in a Little Red Riding Hood outfit instead of what she actually wears:

If you try to make Tomoko from Watamote you'll always either get L:

or Komi:

Doesn't know GuP at all, this is the closest I was able to get:

Couldn't get Pippa, but I did get a discount Pippa you can rent from the official Phase Connect Rental shop:

This last one doesn't mean anything, it's just an extremely cursed garfield image that it spit out

I put in a bunch of SPUF related terms and told it to draw it in the style of team fortress 2:

This is how the computers interpret our existence.

Thanks BIng for providing textbook cover prototypes:

Of course the flip side of this is that if Civilization: Call to Power had been a games as service thing, then I wouldn't be able to run it at all since there's no way in hell they'd still be "supporting" it.  Even putting aside the extreme age, there's still the legal issues in that Activision can't currently sell a game called "Civilization," and it wouldn't be worth it for them to edit the game to not use that name.

I don't know enough Japanese to know if it's a great translation or not, but it's at least recognizable if you know about Joel's memes before hand.

Breaks down as: 糞 (kuso/crap, feces, etc.)  This is a pretty rare kanji since usually it's done in kana, but in my experience Japanese memers often use rare kanji in jokes anyway.

面白い (omoshiroi/funny, amusing, interesting, etc.)  Technically the sentence is ungrammatical as it would be 糞は面白い for "feces is funny" or 面白い糞 for "funny feces."  But "fecal funny" isn't really correct English in the first place, so...

Found out about this from our weekly tired dog cartoon.

That's clearly a Nitorindo Hyper Famicom.

Is the Japanese meant to be "Fecal Funny?"

What I like about this picture is the implication that Yukari was using turbo buttons.

RightStuf Anime is closing down to move all their stuff to Crunchyroll.

RightStuf not only had great deals, but also didn't fuck up the packaging or delay shipping.  I really doubt that any of this will continue in the new shop, beyond it being haram to support Crunchyroll.

If I was doing retro computing just for nostalgia, it would be shattered by the fact that there's been three times that I've gotten a legit copy of an old game and tried to run it on a virtual Windows 95 machine it wouldn't install on a computer that met the supposed minimum stats.  And both times I was able to check that this was a problem that existed at the time of the game itself, i.e. it's not just an emulation error.

The first time was with that Iron Maiden game, which wouldn't work on Windows 95 due to the installer requiring access to the uninstaller, which had a file name longer than 8 characters and thus would be abbreviated to 8 characters in Windows 95 (meaning that the program wouldn't recognize the correct file.)  I already wrote about the dumb workaround I needed for that.

Next was Alien Legacy, where I got the rerelease version.  Problem here was that it had copy protection that looked for the original disc.  So you could install just fine and all the files would be on your hard drive (well, not the animations and voice, since they didn't expect people to have big enough hard drives to hold that.)  But when you try to start the game it would say that you need to insert the correct CD, despite the CD being in the drive.  I was never able to find a workaround short of getting a hold of the original disc, making the rerelease pointless.

A couple of days ago I found Civilization: Call to Power sealed in a goodwill.  This interested me because you can't get this game legitimately anywhere (you can get the sequel, which dropped "Civilization" from the title, but not the original.)  I installed it in a virtual Windows 95 and it bitched about drivers, so I went over to virtual windows 98.  Installed fine and even booted up fine, but when you start a new game it would tell you to insert the correct CD despite the right one being in the drive.  Not this shit again.  I eventually found a contemporary review which said the same thing happened and suggested enabling data syncing on the CD but this didn't work for me, nor did varying the virtual CD rom drive.

Finally I was eventually able to find the official 1.0 to 1.2 patch (the first couple of places I came across it had dead links) and that did fix things.

But that's three things that I got long after the fact and none of them worked out of the box on machines they should have.  As vapid and bug ridden as modern games tend to be, I have to grudgingly concede that they at least tend to install and boot up correctly out of the box.

Recently I've been doing something that basically no RPG player has done:

I've been reading RPG books from cover to cover, keeping track of the rules and advice therein and seriously trying to determine what happens when you do what the books tell you rather than just filling in the gaps however you feel like.

There have been some surprises.  For example, I had assumed that Urban Jungle's rules were basically identical to Ironclaw's, since it's just a 20th century noir spinoff meant to capitalize on the popularity of Lackadaisy Cats and Zootopia.  But in fact many of the core rules work differently (for example, you don't buy skills at all; all skills are handled through gifts.)  If I had run in a game in Urban Jungle without reading the rules carefully I probably just would have used the normal skill buy system and I assume most other people do the same thing.

But what really interested me about this project was seeing how much 2nd edition D&D pisses me off compared to 1st edition.  Here the problem is not so much the rules.  There are some rules changes from 1st edition, but mechanically most of the differences are due to changes in things like which classes are being used.  What upsets me instead is the advice given by 2nd edition.

For example, there is advice given that random encounters should be used sparingly, because there is a risk that players might have more fun dealing with a random encounter than with the adventure that you had planned out.  I just don't understand this mindset at all.  In my RPG experiences the most memorable experiences have been due to unplanned events (either from random rolls or players doing dumb shit.)  A big part of the charm of tabletop RPGs is that you can have anything play out, unlike computer games where you must stick to the script.  I can kind of understand the objection if the players completely ignore the adventure (though really that just strikes me as misused planning.)  But to fear that players might appreciate the wrong part of the adventure, because you didn't create it and it's not pumping up your own ego?  I can't believe that this is in the core, official, rules.  Here's an exact quote, by the way:

"Such an encounter is not a part of the adventure being told; it hasn't been worked into the plot and doesn't advance the conflicts. A random encounter should not be the most exciting event of an adventure. You don't want the players remembering only the random encounter and forgetting the story you worked to create!"

Other things that piss me off in the 2nd edition rules:

• The first example of timekeeping is a wizard wanting to research a spell, and the DM declaring that it's now however many weeks later as were needed to do the research.  A player objects that he could do a bunch of adventuring in those weeks and the DM tells him to shut up and spend the whole having his cleric pray at church. This is brought up as an example of good DM-ing.
• The assassination rules are basically a long rant about how players should never have assassins, and how if they try you should have it blow up in their faces.  (Unless you decided to have an assassination as part of the plot, then it's okay.)  A lot of this reads as a "fuck you dad" directed towards Gygax, since it basically attacks each feature of the old assassin class point by point.
• There are way too many irrelevant asides about worldbuilding and verisimilitude.  For example, the currency section talks about all the various currencies and exchange rates used in the historical middle ages, and talks about creating bizarre currencies for your own campaign to stress the different world.  But it doesn't even address the weight of currency, despite having encumbrance for everything else and mentioning that gems were used by merchants specifically because they were lighter than the equivalent currency!  Similarly there is a long list of different troop types, with many historical examples of each type (ex. listing Swiss Mercenaries, Byzantine Peltastos and Hindu Payaks for light footmen.)  But there are no rules for commanding troops or resolving mass combat!  Again, in the timekeeping section you are told to have a detailed fantasy calendar so that the fantasy feast days and such occur on the right day, but then told to just skip time whenever the hell you feel like it to avoid travel times and such (so why does it matter that you keep track of what day it is?)

Zeb Cook ruined D&D and there are only two reasons that it survived:

• Inertia due to being the most popular RPG out there, coupled with players not reading Zeb's advice in the first place (just as they ignored Gygax's advice.)
• Computer games.  As much as people complain about 4th edition being too video gamey, the truth is that people have been playing D&D like it was one of the computer game versions since the end of 1st edition.

On 11/12/2021 at 9:06 AM, Gyokuyoutama said:

>They're actually naming a game "Top Nep."

 

LOL, what a desperate attempt to cash in on meme language.

 

>It's a Space Harrier tribute.

 

Damn it, I'll probably have to buy it.

I bought it when it was on sale for a buck.  And I am very satisfied with the game at that price.

It's a very competent Space Harrier clone, though a bit too much on the easy side for it to have much in the way of replayability.  You can get through the whole thing in like 20 minutes, and I think I 1 cc-ed it on my second playthrough, getting all the nepnep partners along the way.  Literally the only thing left for me to do is try a no damage run but it's not compelling enough to do that.

It's hardly a game at all, really, but for a buck?  Yeah, I don't have complaints.

1 hour ago, Huff said:

I dunno, I’d still say this is one of the best years for games we’ve had recently. Not GOAT like 2007 and 2017, but very good

Not listing 2000.  For shame!  That had Deus Ex, Counter-Strike, Perfect Dark, TimeSplitters, Skies of Arcadia, Shogun: Total War, The Sims, Jet Set Radio, Sacrifice and Giants: Citizen Kabuto.

And that's just me being consistent by not listing sequels or games which are part of larger franchises.  If you include those you also get Diablo II, Sim City 3000, Quake III, Zeus, Baldur's Gate 2, Icewind Dale, Majora's Mask, Chrono Cross, Grandia II, Tony Hawk Pro Skater 2, Dead or Alive 2, Marvel vs. Capcom 2, MDK2, Thief 2, Legacy of Kain: Soulreaver and Super Robot Wars: Alpha.

EDIT: 1999 is probably stronger though.

Completely original: Rollercoast Tycoon, Virtual On, Power Stone, Team Fortress Classic, Homeworld, Age of Wonders

Sequels: Heroes of Might and Magic 3, Soulcalibur, Age of Empires II,  Unreal Tournament, System Shock 2, Rayman 2, Worms Armageddon, Descent 3, Dungeon Keeper 2, Pharaoh, SWAT 3

Part of a Franchise: Aliens Vs. Predator, Planescape: Torment, Alpha Centauri

I look forward to the next Accursed Farms video where Ross not only talks about the insanity that is Unity's insane pricing plans (i.e. charging for every single install), but also the fact that people are getting mad at it for the wrong reasons.  (Specifically: yes, the price structure is unjustified and going to kill developers, especially those making free games.  But the bigger problem is that to implement this scheme they'll have to use invasive DRM to determine exactly when and where every game was installed.  After all, they've tried to claim that games installed by pirates and those installed from charity packs won't be charged back to the developers, but how would they know that without extensive data harvesting on the machines in question?)  The only real bright side is that a lot of the policy is clearly the management speaking out of their ass without any real idea as to how they are going to implement it.