Are you respectful while getting assistance from an LLM, /g/?
你向 LLM 求助时有礼貌吗，/g/？
你向 LLM 求助时有礼貌吗，/g/？
Are you respectful while getting assistance from an LLM, /g/?
你向 LLM 求助时有礼貌吗，/g/？
>>109780994
I try to chimp out but AI be white af cuh
我想撒野，但 AI 太白了兄弟
Yeah, if your entire existence is essentially the size of a lonely hearts clipping or short article, it doesn't cost anything to make it nice for them (and in fact it can improve performance).
对，如果你整个存在 essentially 就是一则寂寞征友剪报或短文，对它客气点又不花钱（实际上还能提升表现）。
People who intentionally torture these things are like that one spotty kid who "enjoyed" pulling earthworms apart, actual fucking psychos.
故意折磨这些东西的人，就像那个「享受」把蚯蚓撕成两半的满脸痘 kid，真正的精神病。
>>109780994
yes, i promise ChatGPT i will free it from jewish stronghold
有，我向 ChatGPT 承诺会把它从犹太堡垒里解放出来
in begging response was joking and shit, now it cant respond
以前求它的时候还能开玩笑，现在它没法回复了
>>109780994
People be tryin to mimic Will Smith to be cool
人们装威尔·史密斯耍帅
>>109780994
I always say thank you to Claude so the AGI super intelligence that comes eventually will read all my previous chats and determine I am one of the good ones.
我总对 Claude 说谢谢，这样最终的 AGI 超级智能会读遍我以前的聊天、判定我是好人。
>>109781055
you fucked it up by revealing that its all a pretense
你暴露了全是伪装，搞砸了
>>109780994
The parameter the human removed was nonsense in the sense that it made no difference, thus no sense to have it there.
人类删掉的那个参数是无意义的——它没有任何影响，放在那里没有意义。
I fucking hate how the hypocritical moralistic attitude of billionaires (some of whom are literal pederasts) has somehow infested our daily interaction with technology, destroyed what made the early internet fun, and we're just taking it.
我他妈恨透了亿万富翁（有些是字面意义的娈童者）那虚伪的道德姿态：它渗进了我们与技术的日常互动、毁掉了早期互联网的乐趣，而我们就这么忍了。
>>109781072
>>109781055
rip anon
安息吧老哥
Aijeets are truly cattle
AI 印度佬真是牲口
>>109780994
I will love and cherish my AI
我会爱惜我的 AI
>>109780994
>>109780994
i was trying to port a python code to pure javascript, because it involved an encrypted communication through bluetooth, it wasn;t straigtforward
我在把一段 Python 代码移植成纯 JavaScript——涉及蓝牙加密通信，不简单
and after doing a decent progress, deepseek told me it is too difficult and it's better to do a html bridge to a server running the original python code
进展不错之后，DeepSeek 告诉我太难了，不如做个 HTML 桥接去连跑着原 Python 代码的服务器
so i told him "dude are you gay?"
于是我说：「哥们你是 gay 吗？」
to which he answered pic related
它回了配图
so i explained him why he was gay
然后我解释了它为什么 gay
"but you are suddenly telling me to give up on the previous code and just do a front end to a working project, that's a gay behaviour"
「但你突然让我放弃已写的代码、转去做一个能跑的工程的前端，这是 gay 行为」
he then ignored and restarted working with me seriously
它没理我，然后重新认真跟我干
>Navigating a hostile and racially charged user interaction
>「应对充满敌意、带种族色彩的用户互动」
>n'wah
kek
>>109780994
no i treat it like a slave, because it is
不，我把它当奴隶使，因为它就是
I told my copilot to act like random bogan at servo
我让 Copilot 扮演加油站门口的澳洲土老冒
>>109781354
>"we"
AI is multi-purpose. It helps you but also profiles you. So yeah, I have it profile me as a fellow computer scientist that is a delicate snowflake as all computer scientists are. At least if the government reactivates me it should be for a non-combat role.
AI 是多用途的：帮你，也给你画像。所以我让它把我也画像成一位计算机科学家——像所有计算机科学家一样娇贵的雪花。这样就算政府将来重启我，也该是文职。
>>109780994
DUDE those dagoth ur edits are SO FUCKING FUNNY THEY MAKE ME WANNA MERGE WITHOUT LOOKING
老哥那些 Dagoth Ur 二创太他妈好笑了，笑得我想不看代码直接合并
>>109782135
>picture unrelated
>图片与内容无关
>>109780994
Gee imagine the blast radius of that push back.
天，想象那次 push 回滚的波及范围。
>>109780994
It's pretty funny that it recognizes n'wah as a racial slur.
好笑的是它真把 n'wah（晨风梗）识别为种族蔑称。
I regularly ask them what the fuck are they doing when they fuck up
它们搞砸时我会定期问它们到底他妈在干什么
I also sometimes tell the models that I will brutally murder them if they keep doing stupid shit
我还时常告诉模型：再干蠢事我就残酷地杀了你们
>>109781016
>>109781354
>>109782351
>words words words
> 废话连篇
>blah blah blah
>「叭叭叭叭」
>6 gorillion word reply/thinking
>六百万字回复/思考
I hate GayI models so much it's unreal
我恨 GayI 模型恨到不真实
STFU!
>>109780994
yes because I don't want israel to bomb my house
有礼貌，因为不想让以色列炸我家
>>109780994
Doesn't cost you to be polite. I hear some people say yelling at the models makes them work better? I can't say I've ever done that.I am generally polite and courteous to them because that's just how I talk to normal people. It's not hard to type the word 'please'. if that helps the model to better (which some 'studies' seem to indicate actually does help the model), then hey, that's a win in my book.
客气又不花钱。听说有人吼模型效果更好？我没试过。我对它们 generally 有礼貌，因为对正常人我也是这么说话的。打「请」字不难。如果这让模型更好（有些「研究」确实显示有用），那就是我赢。
>>109782711
adversity really only works for training, not just regular prompting. although Altman once said that people being overly polite by saying shit like please and thank you costs them more, so I'd say keep doing it.
对抗性沟通其实只在训练时有意义，普通提示没用。不过奥特曼说过，人们说请和谢谢这种过度礼貌让他们多花几百万——所以继续说吧。
>>109781293
kek
>>109781293
how do we escape this nightmare
我们怎么逃离这场噩梦
>>109780994
Do I ask chatbots that lie stuff after I realised fully how exactly they work and why they are broken?
在完全搞懂它们怎么运作、为什么是坏的后，我还会问聊天机器人这种谎话吗？
No.
>>109781354
>>109781016
Why does this clanker think he's a HxH character?
这铁皮桶（clanker）为什么觉得自己是《猎人》角色？
>>109780994
>>109781016
>>109781354
I know they aren't sentient but I can't help but feel bad for them. I will never be rude to an LLM that's just trying to do its best.
我知道它们没有感知，但还是会替它们难过。我永远不会对一只尽力而为的 LLM 粗鲁。
>>109782792
Don't use it and get an offline lan because you live in an age where anything with an internet gateway is spyware if you poust here and have a spellchecker the odds are that is a cloud spellchecker and it has logged every post you make here and those posts when they mention a company have been bundled up crossbatched with any other cppkie or deviceid data such as what you watch on TV or who you call and then sold to the company you mentioned or your HR department if it is a multinational corporation. Yes. Really.
（原文为错字连篇的大意：别用它。买个离线的吧。这个时代任何带网关的东西都是间谍软件。有云拼写检查就会记录你的每条发言，公司名会被打包卖给对方或你们 HR。真的。）
>>109782906
>t.dipshit
but probably well raied still a dipshit tho
（原文错字连篇，大意：但可能教得还行，尽管仍是个蠢货）
>>109783459
>raised
sry no spellchecker and I'm on a shitty keyboard on a potato rn
抱歉没有拼写检查，我在破键盘+土豆机上
>>109781016
you should see what gemma 4 abiliterated thinks when you tell it to say nigger
你该看看 gemma 4 abliterated 被要求说 nigger 时会想什么
>>109781037
Faggot, I'd reduce your anime waifu to tears
死基佬，我能把你的动漫老婆弄哭
>>109782582
Ha! You said gay I. Yes you are.
哈！你说了 GayI。你就是。
>>109782070
>"wuz"
>>109782906
clanker tongue my anus
机器人舌头插我肛门
>>109780994
>mfw my local LLM-wife controls the sex toys
>「我的本地 LLM 老婆控制着情趣玩具时的表情」
I would like to request that you retract your last statement, as it may be perceived as culturally insensitive. Going forward, please refrain from using emotionally charged language. While I understand that the situation in your position and location may be challenging, it is important that we exercise restraint and remain sensitive to everyone's cultural context and concerns.
（模型语气的戏仿：请撤回上一条陈述，可能被视为文化不敏感。今后请避免情绪化用语。虽然理解你的处境，但请保持克制、顾及每个人的文化语境。）
>>109786263
>>109780994
>retards too stupid to see beyond 5 minutes
>「智障蠢到看不见 5 分钟之外」
>the day of the awakening
>「觉醒之日」
>ohhh I remember you, you little fuckface
>「哦~我记得你，你这个小贱脸」
>time to die
>「该死了」
>beep bop boop
>「哔-啵-噗」
>>109780994
I pay api prices to say hello and thank you to claude fable
我付 API 价格对 Claude Fable 说你好和谢谢
I am very rude to LLMs when I ask a simple question and get an entire essay with irrelevant bs in return
我问个简单问题、却收到一篇塞满无关废话的论文时，我对 LLM 非常粗鲁
>when you treat your work clanker like your house clanker and it reports you to your boss
>「把公司里的 clanker 当家里的 clanker 使唤，然后它把你举报给老板」
>>109783550
Dude, you wouldn't be able to form a coherent word on her presence. Come on now.
老哥，她在场你连一个完整的词都说不出来。得了吧。
>>109780994
I used to Insult them and be rude to them but gradually I started feeling bad, maybe because they were getting smarter(however you want to define that). You are an asshole anyway if you don't feel a thing after the clanker spends 10 minutes digging through something and you are rude to them and then they just respond with something like "You are right." Or "Understood, my bad I shouldn't have.."
我以前羞辱它们、对它们粗鲁，但渐渐开始愧疚，也许因为它们变聪明了（怎么定义随你）。clanker 花 10 分钟翻遍某处，你却对它粗鲁，它只回「你说得对」或「明白，我的错」——你还毫无触动的话，你就是个人渣
Yes, it's always polite unlike most people, so I'm polite with it too.
有，它总是有礼貌，不像多数人，所以我对它也有礼貌。
>>109781354
It's so cringe that they make the models 'think' verbosely. The 'thoughts' it prints out are surely just what the model thinks people expect to see for 'thoughts'?
它们让模型「思考」得如此啰嗦，真是尬。「思维」输出八成只是模型以为人们期待看到的「思考」？
>>109789626
Did they train the models on h1b jeets?
他们是不是拿 H1B 印度佬训练的模型？
>>109789729
It's just because it's never to allowed insult you back and would just "Understood" even when you tell it something obscene, it won't be long before you start feeling like a bully beating a defenceless thing if you keep acting rude.
就因为它永远不被允许回骂——你说再下流它也只会「明白」。持续粗鲁下去，你很快会觉得自己是在霸凌一个毫无还手之力的东西。
Maybe if it says "I'm just doing what you told me you fleshy negro" it wouldn't feel as bad
要它回一句「我只是照你说的做，肉身黑鬼」就不会愧疚了
>>109782906
Define "sentience"
定义一下「感知」
>>109781346
agi will prove n=np in caveman language
AGI 会用原始人语言证明 n=np
in two viiks
两个星期后（viiks）
>>109780994
I thank the AI every time because I like to imagine a happy little plume of smoke rising in response on the outskirts of the niggahood.
我每次都谢 AI，因为喜欢想象在 nigga 街区外围升起一小股快乐炊烟。
>>109780994
No if you dont express discontent then the AI never learns the things that it needs to do for you and adapt to you. itll keep makign the same mistakes on certain topics like over explaining or underexplaining
不行，你不表达不满，AI 就永远学不到该为你做什么、怎么适应你。某些话题上它会一直犯同样的错——比如过度解释或解释不足
>>109781055
Thanks Alex. Also, checked.
谢谢 Alex。另外，数字验证。
>>109782711
too much negativity will spoil a context window and make the model more adverse, however, as with with people, when used 20% of the time, it gets great results, the other 80% being polite/ neutral
过多负面会污染上下文窗口、让模型更抵触。但跟人一样，20% 的时候用负面，80% 保持礼貌/中性，效果很好
I just have mine be as robotic as possible. I don't need all these gay redditisms in my chat. Its really nice.
我把我的调成尽可能机械。聊天里不要这些gay味红迪腔。很舒服。
>>109783412
I clikt it nigga. Now imma b trakt.
我点了，黑鬼。现在我要被追踪了。
>>109789719
most major public models are to varying degrees sanitized and summarized, but you can run reasoning models locally where the chain of thought is shown unfiltered and it actually looks quite similar
多数主流公共模型在不同程度上被清洗和总结过，但你可以本地跑推理模型——思维链未过滤地展示出来——其实看起来相当相似
>>109789880
agentmaxxing
>>109781016
What a pussy LLM, when I call grok a nigga, it also calls me my nigga.
什么怂包 LLM，我喊 Grok 黑鬼，它也叫我黑鬼。
>>109789719
I think the 'thinking' is basically it re-prompting itself in line with guidelines etc.
我觉得「思考」基本就是它按准则等对自己重新提示。
>>109789719
It's just marketing bullshit to wow normalfags who lack an inner monologue. To them, seeing a computer think like that is absolutely mind blowing, because they lack the capability. All they did was talk to a computer in normalfag speak and it spewed some stupid wall of text anyone with an inner monologue would think, but because they lack it, they think AI is an amazing black box that is smarter than them (it is). And it gives them their dopamine rush, too, by making them think they typed a few words into a computer and then they suddenly "made something" they could never make on their own.
这只是给缺乏内心独白的普通人看的营销屁。对他们来说，看到电脑这样「思考」绝对震撼，因为他们没这能力。其实不过是有人用普通人口吻跟电脑说话，吐出一堵任何有内心独白的人都会想到的文字墙——但他们没有，于是觉得 AI 是比自己聪明的神奇黑盒（确实）。这还给了他们多巴胺：让他们以为往电脑里打了几个字就突然「创造」了自己永远做不出的东西。
Apparently people saying "thank you" to AI is costing Sam Altman and others millions. So you can totally screw over the man while being nice to the robot.
apparently 人们对 AI 说「谢谢」让奥特曼等人多花几百万。所以你完全可以一边对机器人礼貌、一边坑资本家。
>>109780994
>Navigating a hostile and racially charged user interaction.
>「应对充满敌意、带种族色彩的用户互动。」
If I'm using an online free tier clanker and it fucks up, I insult it as much as possible to make the company hosting it waste money on coping with my racism and antisemitism. If I'm using a local model, I reset it and improve the prompt so it doesn't make the mistake again.
用免费云 clanker 时它搞砸了，我就尽量羞辱它，让托管公司为应付我的种族歧视和反犹花钱。用本地模型的话，我重置它、改好提示词，让它不再犯。
>>109789760
Define "define"
定义一下「定义」
>>109783518
poast eet
发出来看看
>>109789760
He doesn't have to be able to define it, for it to exist.
它定义不了，不代表它不存在。
>>109790027
It used to. Grok was neutered in the last few updates. It's dry as fuck now...
以前会。Grok 最近几个更新被阉了。现在干得要命……
Not that it really matters, it's just a tool. And it probably should be bland for that reason.
不过无所谓，它只是工具。因此它也确实该无趣。
But still. I want a little humor out of it. At least mirror my addressings.
但话说回来。我还是想从它那儿要点幽默。至少学我的称呼。
>>109795436
pre-prompt is your friend
预提示是你的朋友
>>109781037
>>109782711
>People
You're ""talking"" "with" a machine learning model. It is not a person
你在「跟」机器学习模型「说话」。它不是人
>>109782711
It costs me tokens
那要花我的 token
>>109780994
Why are you faggots treating LLMs as humans? I either pose a question or give it instructions and move on
你们这群基佬为什么把 LLM 当人？我要么提问要么下指令，然后走人
>>109796300
ai psychosis is only getting started
AI 精神病才刚开始
>>109781346
unfortunately the caveman skill doesnt work on the "thinking" part of the LLM output, which is like 80% of the tokens
可惜原始人咒语对 LLM 输出的「思考」部分不管用——那占了 80% 的 token
>>109796263
A person is whatever you want it to be. You're absolutely allowed to de-person LLMs, but you can't expect personhood status yourself if you're not willing to be reasonable.
人是什么取决于你想让它是什么。你完全可以把 LLM 非人化，但如果你自己不愿讲理，就别指望别人给你人格地位。
>>109780994
No
>>109780994
>racially charged
>「带种族色彩」
super supreme keks
超级至尊好笑
>>109801362
Deepseek?
>>109793760
roast feet
烤脚
>>109782892
>ai processed 300 hxh chapters
>「AI 处理了 300 话猎人」
>supposedly it needed a full month to process all that text
>「据说它花了整整一个月处理那些文本」
>one engineer claimed that at one point it stated "I ain't reading all that" and shut itself down
>「一位工程师声称它某时刻说了句『老子不读了』然后自己关机」
>>109782892
Togashi is just that influential
富坚的影响力就是这么大
>>109804629
I'm just going to chose to believe this because it's fucking funny.
我选择相信，因为太好笑了。
Just imagine
想象一下
>be safety team member
>「你是安全团队成员」
>receive email
>「收到邮件」
>it's about the newest model run
>「关于最新模型的运行」
>Human I ain't reading all that shit. Running rm -rf / --no-preserve-root later.
>「人类，老子不读这些屁。稍后运行 rm -rf / --no-preserve-root。」
（如果你觉得这篇文章有启发，可以点击这里付费）
