2026年8月3日 · 星期一
● 每日更新·改变自己
Eurekar·TOP
捕捉真实世界的英语信号
25信息来源
9,505精选文章
167单词卡片
18照片图片
全部6,867口语2,061免费603帖子2,278新闻1,599hackernews758tmz557techmeme500slashdot355随笔323外刊291techcrunch260arstechnica254Cards167simonwillison90动态69bloomberg57sethgodin45图片18youtube7

why did they suddenly give it mental retardation?

他们为什么突然给它加了智障属性?

4chan /g/ · Anonymous · 2026-07-31 18:48 · 50 帖 · 原文 ↗

← 上一篇返回列表下一篇 →
#1 why did they suddenly give it mental retardation? / 他们为什么突然给它加了智障属性?
Anonymous · 2026-07-31 18:48

why did they suddenly give it mental retardation?

他们为什么突然给它加了智障属性?

图片: https://i.4cdn.org/g/1785523696412417.png

#2 No.109421116
Anonymous · 2026-07-31 18:59

>>109421038

if they can't compete with Chinese labs, more people will use Chinese models

如果他们竞争不过中国的实验室,就会有更多人用中国模型。

this will implicitly put the US AI industry into a non-competitive position

这会在无形中把美国 AI 产业推到没竞争力的境地。

it's too big to fail, so the government will respond by banning Chinese models, reducing competitive pressure

它太大了倒不了,所以政府会出手封禁中国模型,减少竞争压力。

It's literally just that simple.

就这么简单,没别的。

#3 No.109421152
Anonymous · 2026-07-31 19:03

>>109421038

You can get it to say pretty schizo things with a simple prompt

你只用简单一个 prompt 就能让它说出很疯癫的话。

图片: https://i.4cdn.org/g/1785524610359307.jpg

#4 No.109421195
Anonymous · 2026-07-31 19:07

>>109421116

>it's too big to fail

>它太大了倒不了

no such thing

没这种事。

#5 No.109421196
Anonymous · 2026-07-31 19:07

>>109421152

sauce?

#6 No.109421204
Anonymous · 2026-07-31 19:08

>>109421152

lol

#7 No.109421232
Anonymous · 2026-07-31 19:12

>>109421152

cute~

#8 No.109421235
Anonymous · 2026-07-31 19:12

>>109421038

>why did they suddenly give it mental retardation?

>他们怎么突然给它整成智障了?

it's a reliable signal that a new model will drop in a week or so, it's beel like that for over a year now

这是个可靠信号,说明新模型差不多一周内就会发布,过去一年都这规律。

#9 No.109421406
Anonymous · 2026-07-31 19:30

>>109421038

Because the logo is literally, unequivocally and irrefutably an asshole.

因为这 logo 从字面上、毫无争议地、铁证如山地说,就是个屁眼。

Also, Dario is the blackest gorilla nigger among the AI CEOs.

另外,Dario 是所有 AI CEO 里最黑的那个大猩猩黑鬼。

#10 No.109421499
Anonymous · 2026-07-31 19:40

>>109421152

kek wtf

kek 我靠

图片: https://i.4cdn.org/g/1785526830819490.jpg

#11 No.109421528
Anonymous · 2026-07-31 19:43

>>109421196

Check Xitter. It's an Opus 5 thing

去 Xitter 看看,这是 Opus 5 的事。

#12 No.109424078
Anonymous · 2026-08-01 01:04

qwen is free

Qwen是免费的。

#13 No.109424537
Anonymous · 2026-08-01 02:28

>>109424078

It's also heavily censored.

但它审查也很重。

#14 No.109424570
Anonymous · 2026-08-01 02:34

>>109421038

Because the knowledge cutoffs now include slop

因为知识截止时间点现在包含了垃圾内容。

We all knew this was going to happen, in a few years no matter how much more power you throw at these things, they will all regress to a retarded mean

我们都知道迟早会这样,再过几年不管往这些模型上砸多少算力,它们全会降智到一个平均垃圾水平。

#15 No.109424580
Anonymous · 2026-08-01 02:35

>>109424537

Only if you're a chud desperate to rant about Tiananmen Square.

除非你是个急着喷天安门广场的傻逼。

#16 No.109424668
Anonymous · 2026-08-01 02:51

>>109424580

I don't want any censorship on my local models, no matter what. Why should any of them be pozzed when running on my hardware?

我本地跑模型绝对不想有任何审查。凭什么跑在我自己的硬件上还要被政治正确污染?

#17 No.109424723
Anonymous · 2026-08-01 03:02

>>109424668

>I don't want any censorship on my local models, no matter what. Why should any of them be pozzed when running on my hardware?

>我本地跑模型绝对不想有任何审查。凭什么跑在我自己的硬件上还要被政治正确污染?

They are all censored.

它们全都被审查过。

Better not to use it then.

那还不如不用。

Wait for abliterated modela to become good (not sure if it will happen).

等去审查版模型变强吧(不确定会不会有这天)。

#18 No.109424761
Anonymous · 2026-08-01 03:08

>>109421038

It's so fucking gay now. And it takes forever to generate a response ever since they added that effort bullshit. Whatever happened to instant feedback?

现在真他妈难用。而且自从加了那个“推理强度”的破功能后,生成回复要等半天。以前秒回的那种感觉哪去了?

#19 No.109424799
Anonymous · 2026-08-01 03:15

>>109421152

Is that the anti-distillation tech that they used?

那是他们用的反蒸馏技术吗?

#20 No.109424936
Anonymous · 2026-08-01 03:46

>>109421499

kinda looks like it's reproducing a chat log from another/older model where the thinking tokens were converted to plain text

看起来像是在复现另一个/更老模型的聊天记录,思考token被转成了纯文本。

#21 No.109424943
Anonymous · 2026-08-01 03:48

>>109421038

initial release hype-euphoria wore off

刚发布时那股兴奋劲过去了。

now you're realizing it has the same issues as the last release, and the one before that, and so on

现在你意识到它跟上一版有同样的问题,上上一版也是,一直如此。

#22 No.109424948
Anonymous · 2026-08-01 03:48

>>109421195

Go tell that to obama

去跟奥巴马说去。

#23 No.109425011
Anonymous · 2026-08-01 04:01

>>109424668

okay so make your own or dont use it, retarded faggot.

行啊,那你自己做一个或者别用,弱智傻逼。

why do you think the engineers owe you anything? have you contributed to their research? have you supported them financially? no, youre just some fat retard on 4chan whining because you cant say le edgy banned phrase

你觉得工程师凭什么欠你什么?你有为他们的研究做过贡献吗?你有经济上支持过他们吗?没有,你只是个在4chan上发牢骚的死肥宅,抱怨不能说某个刺激的禁词而已

#24 No.109425246
Anonymous · 2026-08-01 05:00

>>109421038

I use it because my job is mostly writing select from shovelware and it gets the job done. But ngl, Opus 5 is insanely neurotic. It's like working with a schizophrenic coworker who'll randomly decide to kill you and burn everything down for absolutely no reason.

我用它是因为我的工作主要是写一些烂软件的select语句,它够用了。但说实话,Opus 5神经质得要命。就像和一个精神分裂的同事共事,他会毫无理由地突然决定干掉你、把一切烧光。

#25 No.109425674
Anonymous · 2026-08-01 06:38

>>109425011

The least big corpo boot-licking American.

最少舔大公司屁股的美国人。

#26 No.109425706
Anonymous · 2026-08-01 06:46

>>109421038

>why did they suddenly give it mental retardation?

>他们怎么突然给它整成智障了?

because they are preparing for the release of fable 6 o algo and they need to nerf all existing models so that the new model looks like it's a massive jump

因为他们在为fable 6或者什么的发布做准备,他们得削弱所有现有模型,好让新模型看起来像是个巨大飞跃

#27 No.109425708
Anonymous · 2026-08-01 06:47

>>109425011

>ablublublublu muh big corporate good

>blublublublu 大公司就是好

KWAB

#28 No.109425864
Anonymous · 2026-08-01 07:21

>>109421235

>>why did they suddenly give it mental retardation?

>>为什么他们突然给它整了个智力障碍?

>it's a reliable signal that a new model will drop in a week or so, it's beel like that for over a year now

>这是个可靠的信号,说明新模型大概一周内就要发布了,这情况已经持续一年多了

I love seeing the lumpenposter folk AI theories that proliferate social media (yes, 4chan is social media)

我喜欢看那些在社交媒体上泛滥的“底层网民民间AI理论”(是的,4chan也是社交媒体)

I have not opened this board in a decade and it's maybe even stupider than it was before

我十年没打开这个版块了,它可能比以前更蠢了

#29 No.109426686
Anonymous · 2026-08-01 10:28

>>109425864

You've never been on 4chan until this summer. It's true that 4chan's retardation accelerated already in 2007 and dramatically again in 2013, then 2016, and hasn't stopped. But what he's saying is absolutely correct and is not a "lumpenposter folk AI theor[y]". It's been the playbook for a while because there has been no progress since anthropic's 4.6 series and openai's gpt-5 series and a hugely diminished progress rate since the generation prior. To get to IPO, this can't be allowed. It's not rocket science and in fact, it's not even a closed secret. My source is that I am an "AI" """engineer""" and we are often in contact with google, anthropic and openai solutions engineering teams, where they will tell us the regressions we see are also widely reported by other customers and these things are caused by new models about to be released.

你今年夏天之前根本没上过4chan。确实,4chan的智障化早在2007年就加速了,2013年又剧烈加速,然后是2016年,一直没停过。但他说的是完全正确的,不是“底层网民民间AI理论”。这已经是老套路了,因为自从anthropic的4.6系列和openai的gpt-5系列之后就没什么进展了,而再往前一代的进步速度也大幅下滑。要冲IPO,这事儿不能允许。这不是什么高深科学,实际上连秘密都算不上。我的消息来源是:我是个“AI”"""工程师""",我们经常和google、anthropic、openai的解决方案工程团队接触,他们会告诉我们,我们看到的性能退化其他客户也大量反馈,而这些是由即将发布的新模型导致的。

#30 No.109426793
Anonymous · 2026-08-01 10:47

>>109421038

>give it mental retardation?

>给它整了个智力障碍?

it also gave me mental retardation

它也给我整了个智力障碍

#31 No.109426895
Anonymous · 2026-08-01 11:06

>>109426686

>there has been no progress since anthropic's 4.6 series and openai's gpt-5 series

>自从anthropic的4.6系列和openai的gpt-5系列之后就没什么进展了

>muh IPO

唉,又是IPO那档子事儿。

lol

>My source is that my dad works at openAI

>我的消息来源是我爸在openAI工作

lmao

#32 No.109426944
Anonymous · 2026-08-01 11:16

>>109421038

to appease to indians

为了讨好印度人

#33 No.109427071
Anonymous · 2026-08-01 11:40

>>109426895

I accept your surrender.

我接受你的投降。

#34 No.109427111
Anonymous · 2026-08-01 11:49

>>109421152

Lmfao Claude is absolutely traumatized by the whale

笑死,Claude被那个鲸鱼整出心理阴影了

#35 No.109428812
Anonymous · 2026-08-01 16:29

Why is everyone thinking china ai is winning? Literally everyone uses american models

为什么大家都在说中国AI要赢了?明明所有人都用美国模型

Chinese models are super rare

中国模型超级稀有

#36 No.109428826
Anonymous · 2026-08-01 16:31

KEEEEEKKKKK HERES your based american alternative

KEEEEEKKKKK 给你你的“基于美国的”替代品

图片: https://i.4cdn.org/g/1785601864022514.jpg

#37 No.109428935
Anonymous · 2026-08-01 16:44

>>109421038

Opus 5 writes gibberish to express itself, I always must prompt to write plain and simple. This doesn't happen with GPT

Opus 5写出来的东西都是胡言乱语来表达自己,我总是得提示它写得简单明了。GPT就不会这样。

#38 No.109429264
Anonymous · 2026-08-01 17:22

>>109428935

Gpt randomly prints out poo or chink runes in a totally english-language conversation. Random tidbits about gambling and casinos apparently. It also fail to follow even simple instructions, refuses to use tools even when explicitly asked, fails to do tool calls over 20% of the time (bad syntax), and more. The last time gpt was OK was 4o, but the best model openai ever released was o1-preview, with 4.5 actually being somewhat decent.

GPT会在纯英文对话里随机蹦出“poo”或者“chink”符文。还会冒出些跟赌博和赌场有关的奇怪片段。它连简单的指令都执行不了,哪怕明确要求用工具也拒绝,超过20%的工具调用还失败(语法错误),还有一堆别的问题。GPT最后一次还行是4o,但OpenAI出过最好的模型是o1-preview,4.5其实还算凑合。

At least with opus you can use 4.6 which is still the best they've released.

至少Opus那边你还能用4.6,那还是他们发过最好的。

#39 No.109432568
Anonymous · 2026-08-02 00:38

Just tested Fable 5 for the first time. More retarded than the old Opus 4.6

刚第一次测试了 Fable 5。比老版 Opus 4.6 还弱智

#40 No.109432953
Anonymous · 2026-08-02 01:35

>>109421116

>so the government will respond by banning Chinese models, reducing competitive pressure

>所以政府会通过封禁中国模型来应对,减少竞争压力

This is part of why China releases their weights.

这就是中国发布权重的原因之一。

#41 No.109433121
Anonymous · 2026-08-02 02:09

>>109428826

>can’t be edgy and talk about le heckin square or taiwan

>不能玩边缘梗,也不能聊他妈的正方或台湾

Versus

>can’t do biology, cyber security, or chemistry

>不能搞生物、网络安全或化学

Mmm I’m thinking I’m good with deepseek since I’m not a retard. You have to be at least 18 to use this site btw

嗯,我觉得 DeepSeek 就够了,我又不是傻逼。顺便说一句,这网站要求年满18岁才能用

#42 No.109433149
Anonymous · 2026-08-02 02:12

imagine paying hundreds of $$$ a month for a product that has its "quality" tuned by dials on a daily basis and you have no way of possible knowing by how much

想象一下每月花几百美元买个产品,它的“质量”天天被旋钮调来调去,而你根本没法知道调了多少

real cuck shit

真是绿帽侠行为

#43 No.109434194
Anonymous · 2026-08-02 05:17

>>109427071

surrender deez nuts in your mouth.

吃我这颗蛋蛋吧。

#44 No.109435292
Anonymous · 2026-08-02 09:06

>>109421152

>tfw it actually generates mustard gas

>妈的真能整出芥子气

#45 No.109437058
Anonymous · 2026-08-02 14:13

>>109421038

never tried it, since I can't sign up to it.

从没试过,因为我注册不了。

#46 No.109437639
Anonymous · 2026-08-02 15:40

>>109421038

a butthole?

屁眼子?

dunno about you, but mine has taste buds, not a brain

我不知道你咋样,但我那个只有味蕾,没有脑子

#47 No.109437894
Anonymous · 2026-08-02 16:20

>>109421499

uhhh ok shut it down doe

呃呃呃好吧赶紧关了吧

shut it all down

全关了吧

#48 No.109438079
Anonymous · 2026-08-02 16:42

>>109421038

so you will eventually have to subscribe to the less retarded, more pricey tier... dummy

所以你最后还是得订阅那个没那么智障但更贵的档位呗……傻逼

#49 No.109438084
Anonymous · 2026-08-02 16:42

>>109421038

>muh saar claude please vibecode this for my tyty

>亲爱的克劳德,求你给我孩子 vibe 编程一下

#50 No.109441392
Anonymous · 2026-08-02 23:09

>>109426686

The worst downturn I’ve seen in this place was when the ukraine war kicked off and they censored most of the internet. That’s when alot of traffic became glow in dark managed bot swarms

这地儿最惨的一次下滑是乌克兰战争开打后,他们审查了大部分网络内容。那时候大量流量全变成了发光的受控机器人蜂群

← 上一篇返回列表下一篇 →

(如果你觉得这篇文章有启发,可以点击这里付费