Stabs and Leaves

Era plays games Art blog here

[eralea]

Link to my art tumblr

hi click here for art blog there are watercolour paintings there

so the problem is, I made this primary blog years ago for gaming nonsense and it's floran-themed bc I'm an idiot plant and then later on I made a side blog for posting art and bc I'm an idiot plant I can't figure out how to make it my main blog so we're doing this hackery instead

[eralea]

got these seads from china

I usually get the built-in OCR in Apple Photos to read these manufacturer info tags for me, so that I can put them into a spreadsheet. However, this tag turned out to be the OCR Final Boss, so I'm putting the results here.

Apple Photos: Product Naine. Seads Batch Nurnbor:1995348982 Manufacturar:Wuhanshijlangxiaqhaimaodianzishangwushangxing Address:Hubri Province, Wuhan,Jlangxla D'atrict, Hidden Dr gon laland Sunahine 100 Great Laka Phasa 5, Building 10, Level 1-3, No.7 Manufacturer E-mall:bao|layu8639@foxіnall.com

Google Search: Product Name. Beads Batch Nurnibar: 1995348982 Manufacturer: Wuhanshillangxiaqchaimaodianzishangwushangxing Address: Hubei Province, Wuhan, Jiangxia District, Hidden Dragon laland Sunahin 100 Great Lake Phase 5, Building 10, Level 1-3, No.7 Manufacturer E-mail: bao/layu8639@foxіnall.com

Really, Google? Nurnibar?? What did I do all those recaptchas for?

Microsoft OneNote: Oroduct Name. qeads Batch 995348982 Manufacturer:WuhsnshiJlangxieqt.høirnaodlsnai'hanguuahangxjnp Address:Hube.i Dn gon Island Sunabilu SOC Graat Lake Phan S, Building Løvøl 1-3, Mo.? Manufacturer

Adobe Acrobat: 6rodu_ct Name. ,a ads Batch Numb,r:1995348882 Manutacturar: wu,hanah IJlan91rt114u ~. , maodla nsllha n9\vuaha ng'Xlng Addre••:Hub,&1 Pro"Jnce,Wuhan,Jl•ngxla 1 tH•trlct,Hldden Dn gon lslend a-, 'tDD Gr~at Lake , .,h••• 5, Building 101 L•v•I 1-3, Mo.7 \4anufacrure.-E•mall: baoJlarua&390,oxrn all,. cotn

Microsoft and Adobe are too fucking comfortable up there. How are you guys the industry standards??? Get outttt gtfo die

Dropbox: Product Name. Reads Batch Nurbor:1995348982 Manufacturar:Wuhanshijlangxiaqhaimaodianalshangwushangxing Address:Hubei Province, Wuhan,Jiangxla Diatrict, HIdden Dr gon laland Sunahine 100 Great Lake Phase 5, Building 10, Level 1-3, No.7 Manufacturer E-mall:bao|layu8639@foxіnaіl.com

DeepL: Product Naine. Seads Batch Nurnbor: 1995348982 Manufacturar: Wuhanshijlangxiaquhaimaod/anzishangwushangxing Address:Hubei Province, Wuhan,Jlangxla D'atrict, HIdden Dn gon laland Sunahine 100 Great Lake Phase 5, Building 10, Leval 1-3, No.7 Manufacturer E-mall:bao|layu8639@foxіnall.com

This is a translation app, but I remembered it could read images.

I also remembered that the AI chatbots can read images, so I threw it at them too, just to be fair.

ChatGPT (GPT-4-turbo): Here is the extracted text from the image: Product Name: Beads Batch Number: 1895348882 Manufacturer: Wuhanshijiangxiequaimacdianzi shangwushengxing Address: Hubei Province, Wuhan, Jiangxia District, Hidden Dragon Island 100 Great Lake Phase 5, Building 10, Level 1-3, No.7 Manufacturer E-mail: baoflayu8633@foxіnall.com

Google Gemini: Sure, here's the OCR output from the tag: Product Name: Beads Batch Number: 1995348982 Manufacturer: Wuhanshillangxieqhaimaodianzishangwushangxing Address: Hubei Province, Wuhan, Jiangxia District, Hidden Dragon Island Sunshine 100 Great Lake Phase 5, Building 10, Level 1-3, No.7 Manufacturer E-mail: [email address removed]

Interestingly, it's a little different from Google Image Search's OCR. It also cut out the email, probably to reduce abuse, but it's not like anyone so far has managed to decipher it.

Claude 3.7 Sonnet: I can extract the text from this image for you: Product Name: Beads Batch Number: 1995346982 Manufacturer: Wuhanshijiangxiaquhaimeodianzichangwushangxing Address: Hubei Province, Wuhan, Jiangxia District, Hidden Dragon Island Sunshine 100 Great Lake Phase 5, Building 10, Level 1-3, No.7 Manufacturer E-mail: baojіayu8639@foxmaіl.com

So close, small mistakes in the batch number and manufacturer name

This is the only western tech to correctly identify the email provider. None of the others so far could do it. In fact, Apple's OCR often renders qq.com addresses as "aq.com", "gq.com", "a9.com" etc. It's 2025, China is the world's manufacturer, and your training data doesn't include Chinese email addresses? Westerners' biases will be their own downfall.

DeepSeek-V3: The text in the image appears to be in Chinese.

啥

The text in the image appears to be in Chinese. Here is the OCR result: 产品名称: 9eads 批号: 1995348982 生产商: 武汉市江夏区藏龙岛粟岸湖 100 大湖第5期10栋1-3层7号 生产商邮箱: baojlayu8639@foxmaіl.com This corresponds to the following in English: Product Name: 9eads Batch Number: 1995348982 Manufacturer: Wuhan City, Jiangxia District, Hidden Dragon Island, Su'an Lake 100 Great Lake Phase 5, Building 10, Levels 1-3, No. 7 Manufacturer Email: baojlayu8639@foxmaіl.com

DeepSeek is doing some extremely demented 11th-dimension logic here — did it correctly translate the actual address of this business, or did it hallucinate a similar but unrelated address? Googling the Chinese address returns a residential complex called 阳光100大湖第, which does translate to Sunshine 100 Great Lake... but it's not a factory or office building, and doesn't look like a place where one would manufacture 9eads.

At any rate, it's either completely wrong or has gone above and beyond to doxx this manufacturer, and I can't tell which one has happened.

Qwen 2.5-MAX: Sure, here is the OCR text from the image: Product Name: Beads Batch Number: 1995348982 Manufacturer: Wuhanshijiangxieqiaimaodianzishangwushangxing Address: Hubei Province, Wuhan, Jiangxia District, Hidden Dragon Island Sunshine 100 Great Lake Phase 5, Building 10, Level 1-3, No.7 Manufacturer E-mail: baojіayu8639@foxmaіl.com

It's still slightly wrong, but extremely close. I think the company name is quhaimao (which Qwen read as qiaimao, and Claude read as quhaimeo), but this is as close as we're gonna get.

Edit: I have a guess for why Qwen was the most accurate. The parent company is Alibaba, so they must have underlying tech that can scan and read billions of these tiny, poorly-translated, poorly-printed tags. If there wasn't a gap and smear in the company name, it would have gotten it 100% correct.

Amazon's OCR tech can probably also do it, but I don't wanna sign up for AWS.

unrelated trivia: when I started this spreadsheet I found a lot of manufacturers with "dianzishangwushangxing" at the end of their names. Googling this returned nothing, of course. Then I remembered I could just type it into my pinyin keyboard...

电子商务上行

"e-commerce startup", I think?

[mothercain]

like yall cannot be stupid enough to scream fuck israel all year (rightfully so) and then go happily vote for kamala and meme the fuck out of her like everybody did trump like she didn’t just get up on that debate stage and say israel has the right to blow the entire levant to smithereens, do notttttt piss me off

[eralea]

CHAPPELL ROAN IS RIGHT

AND YOU’RE RIGHT TOO 💗

[eralea]

WW2 story

My grandmother often told us this story about WW2: During the Japanese invasion, she disguised herself as a man and camped out in the forests.

Our forests are home to the Malayan Sun Bear, and Malayan Tiger, and all kinds of crocodiles, and apes, and monkeys, and SOOOO many snakes...

But, as a young lady in her late teens, camping out with all the wildlife throughout the years of occupation was far safer than getting kidnapped by Japanese soldiers.

[eralea]

Infinite Craft and AI Ethics

There's been a big quarrel on twitter about Infinite Craft being AI generated slop, and a lot of people are arguing that it doesn't plagiarise or steal jobs so it "doesn't count as AI generated slop".

But the issue in this particular case isn't copyrighted data, it's biased data.

Maybe it's because most of the players are on the younger side, so they don't remember the debates about unethical AI regurgitating harmful outputs, caused by training data containing unfortunate biases.

So here's some AI slop

I know a stupid browser game is the least of Palestine's issues right now. If they're not being killed by bombs and invading settlers, they're dying from starvation and food poisoning. The zionists have killed entire families, over 15000 babies and children, and the current death toll is almost certainly undercounted. We won't know the full damage of the Second Nakba for months, if not years from now.

So in the grand scheme of things, an AI slop game that says "Palestine + Child = Terrorist" is just a minor thing, a little accident caused by an algorithm parroting the racist narratives told by government propaganda worldwide. It's not the developer's fault that the data contains this bias.

But I can't shrug it off, can I?

There are people with 🇵🇸🍉🔻 in their names and profiles saying the game is "just harmless fun". I can't stand it. This isn't harmless. This is the racism the zionists use to justify their genocidal mania. This is the bigotry that causes most people to react to the genocide with apathy or even glee. This is the islamophobia that is so ingrained in our media, in our whole society, that our AI systems will casually reflect it back at us and most people wouldn't bat an eye.

And I thought the game was fun, too. I don't like generative AI, but I thought this word association algorithm seemed like a perfectly fine use of regular AI. Then I tried to make "Free Palestine" and kept finding "terrorists" instead.

It was a harsh reminder that these generative systems are not random or neutral. And I knew it, I had read Timnit Gebru's writings on racial and anti-black bias in AI, I have a collection of ridiculous whitewashed AI pics, I knew that AI had a racism problem but I just wanted to play the silly word game.

I think a game like this can still be done, but it needs human involvement to prevent "Ooops! Accidental Zionist Propaganda: The Game". The lone dev can't pre-emptively block all the possible abuse cases. He doesn't post about politics so I have no idea if he cares about AI ethics (or genocide) enough to fix it, and he probably won't see my complaints anyway.

And it's not even really about the game. It's about AI inheriting and reinforcing the worst parts of society. The AI model the game is built on (LLaMA) associates Palestinians with terrorism. Will other technologies built on this model return the same output? Are there other nasty biases in the data that we haven't found yet, biases that will be perpetuated every time someone develops anything using this model?

[eralea]

a dilemma

A small part of me thinks I should uncheck this and upload thousands of nightshaded images

[eralea]

did i really upload this to my art blog and not my gaming blog

[eralea]

my least favourite thing about Nikki is the way she sends no less than THREE notifications and it’s not even stam time yet bc server time is only 18:00 rn

[eralea]

Big thumbnail ad link for AI-generated adoptables next to my artwork, how do I set it on fire?

This kind of thing makes me think I should get out of deviantArt too but I can't remove myself from so many platforms.

I stopped promoting myself on twt because the policy now makes clear that you agree that anything you upload will be scraped to train AI.

(People say it's only for training algorithms, but given that Elon regularly comes up with stupid unpredictable new ideas, and given that he has directly stated that he wants to make a "non-politically-correct" gAI to compete with OpenAI, I just don't buy that and Idk how anyone can trust that they will only ever use it for recommendation algos.)

At least there's a checkbox to say I don't agree to be scraped when uploading to dA (goodness knows how effective it actually is).

[lunar-wandering]

fun fact!! it turns out that now when u make a new blog, tumblr forces you to follow 3-4 people before you can change your icon or modify your blog in any way!! this, of course, means that, yes, some of the "potential bots" many of us have been automatically blocking could have possibly been genuine new users who were only just seconds in to having an account!!! tumblr is literally screwing new users over!!!!

[eralea]

OHHHH that explains so many weird notifs

I'm so relieved that it's "newbies stumbling around bad UX" and not "invasion of the weirdly silent porn bots"

[eralea]

6 years of Connous Beach event

and I'm only realising now that the four Fun Fun Cart employees are called Doug, Ice, Cowan and Lemon.

Lemon Ice and Corn Dog.

Fking hell Mabi how dare you inflict this cringe on me

[eralea]

zoey waifu fr ❤️❤️❤️😻

[eralea]

[eralea]

awww this auto-dress is so sweet

I wouldn’t mind if a date turned up like that

the other one tho

uhhhhhhhhhhhhhhhh

[eralea]

this furniture gacha is SOO CUUUTE

[eralea]

rice gege pls for 33 bucks i can get a full ceramic dinnerware set off aliexpress

[eralea]

I like this game bc you would look at this and think “team is being carried by lv80 Feilong and Donkey” but actually no, team is carried by lv40 Dongpo cleaning poison every turn bc they die by round 5 otherwise