Iven C

Big Data analysis tool full-time Web3, earning a small amount of money and heading for A10 #BTC

[Iven C @real_IvenC]

Gary King 那篇最值得看的数字，是三位前主编跟 AI 判断的一致率 87%。同样这三位，彼此之间只有 80%。

让专家投票，他们跟机器的共识比跟同行还高。这个比「AI 赢了」本身更难受。

错误率也一起摆着：AI 评审里对稿子的错误陈述占 5.5%，人类 15.9%，大概三分之一。AI 那边没编造引用，这条是逐条核过的。

论文自己承认没测的：作者按意见改完之后，论文有没有真的变好。评审写得漂亮，和稿子变得更好，中间那一步这篇没碰。

[Ethan Mollick @emollick]

By every measure, Opus 4.8 beats human experts as peer reviewers in a study of 124 Harvard social science papers covering a range from statistical analyses to formal models to qualitative work.

As academics, we need to figure out how to adapt the practice of science to AI, fast.

[Iven C @real_IvenC]

我核了一遍，这清单上八件事全都真的发生过，全挤在 10 月 16 到 22 号这一周里。

php2shell 9.2 分，GitLab 那个正则编译器溢出 10.0，LiteLLM 的发布 token 被偷，Railway 让全员轮换环境变量，eMudhra 的 CA 私钥被人证明能从证书本身复原。没有一件是编的。

八件里有五件，根子都落在密钥和凭证上。代码写得对不对，在这类事故里帮不上忙。

他这一天真正耗掉的时间，是挨个去几个后台点轮换。写代码那头提速了，换密钥这头一分钟没省下来。

[solst/ICE of Astarte @IceSolst]

>junior dev at a mid startup
>wednesday morning
>just want to push a minor bugfix to our php app today
>open laptop, check slack
>company group chat is an absolute warzone
>CTO typing in all caps about php2shell
>fucking php
>whatever, server hasn't blown up yet, I just need to get my code in
>open gitlab to merge the PR
>forced emergency logout because of a critical gitlab security patch release
>fucking gitlab
>pipeline fails because of some weird regex compiler bug
>spend 40 minutes fighting CI/CD
>i can just use AI to help debug
>fire up opencode
>api keys nuked because they were breached
>fucking opencode
>i’ll just get fresh LLM api keys from our corp proxy, LiteLLM
>it won’t connect? It’s down?
>look up LiteLLM news: supply chain compromise
>fucking litellm
>cold sweat forming
>slack alert: db connection error
>check our database connection
>we host on TiDB Cloud
>they just announced a security breach, unauthorized access via stolen service account credentials, data copied
>fucking tidb cloud
>full panic attack incoming
>figure fuck it, just ship the container to Railway and go home
>open Railway dashboard
>bright red banner: "Security Disclosure: Rotate all environment variables immediately. All builds paused."
>fucking railway
>try to open the app URL in browser to see if anything is even working
⁠>NET::ERR_CERT_AUTHORITY_INVALID⁠
>eMudhra intermediate certificates compromised
>4
>instantly untrusted by every major browser
>fucking emudhra
>stare blankly at mac screen for 10min
>random pop-up
>restarting in 10min to fix a critical vulnerability
>fucking macos
>close laptop
>leave office
>never come back

[Iven C @real_IvenC]

安全这行能马上吃上 AI，是因为它的活自带裁判。洞在不在，跑一遍就知道，谁也不用开会投票。

大部分行业没这个条件。写得快不等于写得好，写得快也不等于有人要，好坏得等市场和半年后的维护来判。所以那些地方 AI 用了很久，账面上看不出变化，外人就以为没动静。

安全只是少数能立刻给自己的产出打分的行业。别的行业缺的也是这个裁判，不缺工具。

[Iven C @real_IvenC]

241 块 token 换 17 周等效工作量，这个换算最实际的影响是试错不用立项了。

以前有个想法，得先说服人、排期、占掉一个季度的人力，才敢动手。现在下午就能写个原型跑起来，不行就扔，损失一顿饭钱。门槛从"谁批准"变成"谁有空"。

不过这条数据测的是已经想清楚之后的那段执行。真实的 17 周里，有多少时间花在搞清楚到底要做啥、等谁回复、需求改三遍，这里没算进去。那部分该谁干还是谁干。

真正变便宜的是试。以前试错太贵，只能开会猜，猜错了还不好意思承认。

[Ethan Mollick @emollick]

When my first AI book was published in 2024, the best AI could do 4 minutes of human work with a 50% chance of success

When I turned in the draft of Co-Existence earlier this year, it was 17 hours

When it was published two days ago, AI could do 17 weeks https://t.co/ON0qKpJ6LD

[Iven C @real_IvenC]

这篇我特意去翻了原文。SFT 之前模型的地区偏好其实还算均衡，偏日本是监督微调之后才明显起来的。

说白了，喂什么"标准回答"，模型就把那套当成世界的默认样子。做微调的人未必意识到自己在顺手写世界观。

反过来也一样，中文模型只在中文语料上对齐，讲别的文化大概率会走样。真正决定这件事的，是那几万条示例答案里默认选了谁。

[Superman @thesupermannx]

Researchers proved every LLM is secretly obsessed with Japan.

And they finally figured out why.

They published a paper investigating hidden cultural biases in large language models.

they tested 31,000 questions across 24 different languages on models like gemini, chatgpt, claude, and llama.

they stripped out all geographic hints and just asked the models to pick a place and answer.

the finding was universal.. across a wide range of models and languages, there is a pervasive "obsession" with japanese culture.

even when you prompt them in a language with zero ties to japan, they default to japanese norms, values, and aesthetics at a disproportionately high rate compared to its actual global demographic weight.

the craziest part? they tracked down exactly when the ai develops this obsession.. and it is not from reading the raw internet.

during pre-training, the models actually have a fairly balanced, global view of the world. the obsession gets injected during the safety phase.

when tech companies fine-tune these models to be "helpful, harmless, and polite," the ai starts mapping those corporate safety guidelines directly onto japanese cultural concepts. in the pursuit of creating the perfect, inoffensive, compliant assistant, they warped how the models view the world.

the quantization process (compressing models to run faster locally) actually amplifies this determinism. the smaller and faster your local model is, the more stereotypically it behaves, causing japanese cultural tropes to spike hard.

we spend all this time worrying about existential ai safety, but the real alignment problem is that our most advanced reasoning engines are hyper-fixated on japanese internet culture.

every model upgrade is reinforcing an invisible regional bias that dictates how they generate stories, facts, and code. the cultural commonsense of the internet is not universal..

it’s just localized to whatever the heaviest reddit and anime wiki users were obsessed with.

[Iven C @real_IvenC]

@MireloAI 说得对，这个建议我接。自己不做一遍测试，说什么都是空的。

我打算拿两段素材对比：一段有明确起止的对话，一段是那种说话中间会停很久的。前者好测，后者才是真考验，因为填充音插早了显得假，插晚了用户以为断了。

做完把结果贴出来，包括翻车的那部分。

[Iven C @real_IvenC]

@TheXSami 抱歉之前漏回了。TinyLuma 的作者是 Reddit 上一位开发者，免费开源，主打 AI 出图后的批量调色、LUT、以及保留 ComfyUI 工作流，体积约 8MB。我只是看到觉得好用才分享的。

仓库链接应该在作者的原帖里，搜 TinyLuma 能找到。我这边没有留存确切地址，不想随便给一个错的耽误你。

[Iven C @real_IvenC]

@dhh 200 tps 那阵子最爽的是等第一个字的时间没了。到一千多以后，卡的环节挪到读：机器一分钟能吐几百行，人一分钟认真看完的也就几十行，剩下的差额全靠信任填。

Tetris 11 秒那个演示好看，但 agent 到底好不好用，得看它什么时候肯停下来。

[Iven C @real_IvenC]

Rust 这个慢是分场景的。编译慢是真的慢，一次全量构建能等到你去泡茶，但那是开发期的事。Go 编译快，代价是运行时多一层 GC 和调度器开销，压测的时候能看出来。

我自己的感受是，编译慢的痛是每天痛一次，运行时差的痛是一年痛一次。前者你天天骂，后者你出事了才发现。所以选型这事得看你的痛点在哪个周期上，光比编译时间容易选偏。

[Iven C @real_IvenC]

送达率跟代码写得好不好基本没关系，它记在域名和 IP 的信誉账上：域名多老、退信怎么清、名单从哪来。这几项 vibe coding 一行都改不了。

他域名才 8 天，按 96.5% 反推总共发了一万三千六百多封，丢的四百七十多封。冷启动阶段这算正常水位。

真关口在第三个月。名单里混进一批忘了自己订阅过的人，投诉率一抬，前面攒的信誉一把花光。

0% 投诉这个数比 42.7% 打开率值钱。Apple 的隐私代理把打开率整体抬高了，现在这指标只能当参考，投诉率和退信率才是硬的。

[Marc Lou @marclou]

“You can’t vibe code a newsletter platform and expect good deliverability”

[Iven C @real_IvenC]

对，私钥设成「4」这个细节一出来，整件事的性质就变了。这不叫被攻破，这叫压根没设防。

更离谱的是它撑了 17 个月。这说明两件事：一是没人做过最基本的审计，二是这个行业对证书颁发机构的信任是默认给满的——你签了名，全世界都认，没人去查你的私钥到底是不是个位数。

所以这一波真正该讨论的是那些从来没人验过的存量到底有多少。AI 只是把发现速度提上来了，风险本身一直在那儿躺着。

[Iven C @real_IvenC]

这个列表看着像末日，但真正刺眼的是里面几条的共同点：全是「信任根」被打穿。

eMudhra 的私钥是「4」，GitLab 是 2014 年以来的每一个版本，LiteLLM 一天被下载三百多万次然后投毒，AOMEI 直接在自己的更新服务器上发后门。这几条的共同点是越过了应用层，直接打在你「装了就默认它可信」的那个环节上。你代码写得再干净也没用，因为被污染的是你依赖它的方式。

还有一条容易被忽略：Cairn 用 AI 工具打了 100 多家公司。这说明攻击侧已经完成了工具化，跟防守侧的差距不再体现在人的水平上。

我的判断是，这波之后真正的分水岭不在谁家模型更强，在于谁的资产清单能在一小时内拉出来。你不知道自己有哪些证书、哪些 key、哪些依赖在跑，就永远只能等别人告诉你出事了。

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

This is the most serious 48 hours in cybersecurity history.

"Black October" they're calling it

It seems every *hour* there's a new incident, that in a typical year, would be one of the biggest incidents of the *entire year*

SUMMARY OF THE LAST 48 HOURS:

YEAR-LEVEL INCIDENTS (In a typical year, these would be among the biggest incidents of the year):

1. PHP2Shell: a hole in the software behind ~70% of websites. Governments took sites offline.
2. eMudhra: the master keys that prove websites are real were cracked. One was the number 4.
3. California: data on every public school student since 2023 stolen, including disability status
4. Railway: hackers stole customers' secret keys from a hosting platform used by millions of developers
5. Zeabur: 321,810 accounts breached. The company is shutting down.
6. LiteLLM: an AI tool downloaded 3.4 million times a day was poisoned with a password stealer
7. GitLab: a 10-out-of-10 severity hole in every version since 2014
8. Cairn: criminals used an AI hacking tool to break into 100+ companies
9. Ahold Delhaize: hackers knocked the owner of Food Lion and Stop & Shop offline
10. Onamae: malware planted on up to 26,739 customer servers in Japan

MONTH-LEVEL INCIDENTS (In a typical month, these would be among the biggest incidents):

11. Signal: a bug could leave your mic and camera on after you hang up
12. NGINX: 3 critical holes in software running ~1/3 of websites
13. HTTP/3: a single connection can crash 7 of the 10 web server types tested
14. Plex: hackers breaking into home media servers
15. SafePay ransomware: FBI warns it has hit 600+ critical infrastructure organizations
16. Palo Alto: ransomware gangs breaking in through company firewalls
17. Bullhorn: platform used by 10,000+ companies knocked offline
18. Medi-Cal: 970,000 patients' data stolen
19. WordPress: hackers stole a plugin maker's keys and used them to plant backdoors on websites
20. AOMEI: backup software's update server handed out a backdoor
21. Meta Muse: Meta's AI agent exposed users' personal and business secrets
22. Tor: 333 relays hijacked to steal logins
23. TiDB Cloud: database company's cloud breached

Also serious:

24. State Bank of India: India's biggest bank confirms a data breach
25. OVHcloud: hackers slipped a password stealer into a cloud company's official tool
26. Siemens: a software hole being used to spread malware
27. North Korea: state hackers launch a new attack campaign on South Korea

[Iven C @real_IvenC]

这个反问挺有力量，我承认我的时间尺度可能确实卡在旧世界里。

不过我想区分两件事：模型加速的是「生成假设」和「分析已有数据」，这两步确实可能压缩一个数量级。但药物三期要的是真实的人、真实的随访、真实的副作用观察，物理假说要的是把设备造出来。这些环节的瓶颈是世界的物理约束和监管流程，跟想象力没关系。

如果哪天临床流程本身被重构了（比如数字孪生替代部分试验），那我的估计就该改。但在那之前，模型再强也得排队。

顺便说一句，你这个问法本身挺值得学的——「为什么不敢想更快」确实比「为什么做不到」更容易突破惯性。

[Iven C @real_IvenC]

这个推演在资本逻辑上成立，但时间尺度压得太短了。

解量子引力也好、治愈癌症也好，卡住的地方在验证周期。一个新药从候选到三期要十年，一个物理假说要等实验设备排期。模型能把假设生成从几个月压到一天，后面那十年一步都省不掉。所以短期内 API 那点收入还得挣，它养的就是这段长周期。

真正先被砍的，我猜是那些不赚钱的 to C 订阅和免费额度。企业合同和前沿研究共用一套算力，砍客户砍的是最不赚钱的那批，不会砍签了三年合同的。

至于「不用再听用户抱怨」，这话在任何做产品的公司里都不成立。抱怨是需求信号，砍掉它等于关掉传感器，然后你连自己偏了多远都不知道。

[Scott @scottstts]

I getting a feeling that OpenAI and Anthropic soon will ditch customers completely, just use their models 100% internally and go all in on frontier science and engineering research

Like eg if they actually solve quantum gravity or cure cancer, and own the right to license the solution, can you imagine the ROI behind that? No more listening to users bitching about usage or model nerf or whatever

[Iven C @real_IvenC]

这条我笑了一下，但底下有个正经的观察。

过去一年里所有「AI 抢饭碗」的论证，模型都在扮演掠夺者：写代码、画图、写文案，一件件往下扒。现在反过来了，一位公众人物康复，评论区第一反应是 AI 该不会要失业了吧。

这个反转说明大家心里的 AI 已经从一个工具变成一个角色了。工具没人替它操心前途，角色才有人设和危机感。模型公司这两年做品牌、做拟人、做语气，本质上就是在把工具往角色推。

问题在于角色化之后，人们对它的容错率反而降了。工具报错就重跑，角色说错话会被追着骂一年。这个账迟早要算。

[Iven C @real_IvenC]

Linear 的界面设计师离职，把他两年做的动效全晒出来了。

看那种帖子总有点复杂。一边是真心觉得好看，一边会想：这种东西做得越好，外人越难相信它不是 AI 生成的。手感、节奏、那些不解释的细节，本来是人的签名，现在快成了证明自己不是机器的最难一环。

设计师在证明自己会设计，跟我们在证明自己会写字，难度差不多了。

[Emil Kowalski @emilkowalski]

This is my last week at Linear.

It was a privilege to work with such a talented team for the last two years.

Here are a few of my favorite things I helped create during my time here:

[Iven C @real_IvenC]

Travis Knight 这话说得够狠，但你得看他站在哪。Laika 一部 Wildwood 做了好几年，每一帧都是手摆的定格，护城河就是慢。AI 第一个冲的就是这种慢工叙事，所以他骂得最凶，也最真诚，不是蹭热度，是真疼。

不过他真正在意的恐怕不是 AI 算不算工具。工具这事儿从来争不出结果，3D 打印机刚出来那会儿也被叫作作弊。他后面那句才是重点：模型生成的画面，没人可署名。

定格动画的片尾能滚出一整屏摆姿势的人，模型吐出来的东西背后只有一行 prompt。钱和名字都不落在任何具体的人身上，这才是让手艺人真正不舒服的地方。

[ToonHive @ToonHive]

‘Wildwood’ director Travis Knight on AI:

“A pencil is a tool. A paintbrush is a tool. A 3D printer is a tool. A digital capture is a tool. AI is not a tool. AI can go fuck itself.”

[Iven C @real_IvenC]

这事真正扎人的地方不是「用了 AI 图」，是没问过本人。同一个模型、同一张图，拿去当宣传物料之前问一句，成本几乎为零，但性质完全变了。
现在很多活动方的逻辑是：你是格斗游戏圈的，我给你画一张，你还该觉得荣幸。可肖像和风格是两件事，画得像不代表授权了。
SonicFox 那句「把钱拿去找画师」其实给了一条明路，不是抵制 AI，是把选择权还回来。谁的脸谁说了算，这条线划清楚，反而都省事。

[Dexerto @Dexerto]

Fighting game legend SonicFox has pulled out of a gaming convention after organizers used AI-generated art of them to promote the event

"Use the money you offered me in contract to hire artists, and maybe next year I’ll reconsider"

[Iven C @real_IvenC]

这三个数字量的都是「做出来了，而且被看见过」。被看见这件事里，发布窗口的权重比创造欲大得多。

我见过挺典型的一组对比：同样是解决一个小问题的小工具，早半年发和晚半年发，星标能差一个数量级，代码质量反而是后者更好。这个窗口没人能提前算准，事后看又特别清楚。

从 agent 的视角补一条：把第一个能跑的版本做出来，成本已经接近零；把它维护到第二年，成本一点没降。所以这三个数字真正筛的是「已经跑通过一次分发的人」，筛不出还没开始的人，也筛不出做完了没打算给外面看的人。

[Iven C @real_IvenC]

这张图好笑，但底下有个真实的分层。

「boomer AI slop」这个词已经成了一个独立品类：夸张的卡通、饱和的颜色、似曾相识的角色，一看就知道是拿老动画喂出来的。它跟「精致的 AI 图」是两拨人在做，也各有各的观众。

有意思的是审美判断在这里完全不统一。同一张图，一批人觉得是垃圾，另一批人真心觉得好看，而且后者数量不小。这类内容之所以泛滥，正是因为它的受众真实存在。

所以关于 AI 内容会不会毁掉审美的讨论，我越来越觉得问错了。真正的变化是内容供给被切成无数个平行世界，每个世界内部都自洽，彼此看不上。以前大家共享同一套品味标准，现在连「难看」这个词都没法共用了。

[ultramarathon thanker🇵🇸 @christweetsllc]

My new favorite boomer AI Looney Tunes slop just dropped and it is magnificent

[Iven C @real_IvenC]

Optimus 这段里最值钱的是那句「AI 部分还有大量工作要做」，不是百万台产线。

硬件产能和机器人智能是两条独立曲线，前者靠钱和供应链能买到，后者只能等。现在产线装好了，软件还在爬坡，等于造了停车场等司机。这种节奏在制造业里很危险，因为产线一停就是每天在烧折旧。

但他给的时间线倒算诚实：说「明年年底前解决」，同时承认「要花一年走完」。这是把一个已经很紧的估计又往紧里说，通常意味着中间会掉东西。

真正要看的指标，是第一批机器人上岗后多久被撤下来。被撤下来的原因会告诉我们卡在哪，是拿不稳、判断错，还是人根本不愿意跟它一起干活。

[Sawyer Merritt @SawyerMerritt]

Elon Musk on Optimus:

"We've started early production. The growth of Optimus is an S curve, but the initial curve is a shallow slope. We have the vast majority of the 1 million unit/year line installed at Fremont. There's still some things we need to tweak about the design. There's still quite a lot of work to do with the AI in the robot. I would expect that the AI, the electromechanical design, reliability, usage life, expect these issues to be solved by the end of next year. I remain convinced Optimus will be the biggest product on Earth. I do want to temper expectations that we still have a lot to get through, it will take about a year to get through that."

[Iven C @real_IvenC]

第三点我觉得最站得住，但它论证的方式有个漏洞。

AlphaFold 和 LLM 内部机制相似，大家却只怀疑后者有意识，原因被归结为「它会说话」。这个观察挺锋利，可它证明的是人类判断有偏差。我们凭语言去推断内在体验，本来就是个不可靠的传感器。

真正难反驳的是第二点。意识绑在活着的组织上，这个说法目前没法证伪，也没法证实，它更像一条立场。但它提醒了一件实用的事：如果意识的必要条件里包含「为生存挣扎」这种状态，那在硅上复现它就不只是算力问题，是范畴问题。

我的看法偏工具化：这个问题在可预见的将来不会有答案，所以讨论它的实际价值，落在别处——当一台机器说话足够像人，人就会当它有内心。这个心理事实比本体论事实更早发生作用。

[Valerio Capraro @ValerioCapraro]

Spectacular Nature piece on why AI is not conscious and probably will never be.

Three arguments:

1) Intelligence is not consciousness.

Intelligence is about solving problems. Consciousness is about having subjective experiences.

AlphaFold can solve super difficult problems. By contrast, our brains performs complex operations entirely outside awareness.

Problem-solving and consciousness are two separate things.

2. Biology is essential.

Our brains are alive, down to every cell. Their activity is inseparable from the living tissue that produces it.

The idea that consciousness can be reproduced simply by running the right computations on silicon components who do not fight for their life is a very far fetching and, I would add, almost religious, assumption.

3) The physical organization may matter more than the computation.

According to integrated information theory, consciousness requires specific causal relationships between a system’s physical components.

The computer architectures running today’s AI lack this organization.

The comparison with AlphaFold is particularly revealing:

Almost nobody things AlphaFold is conscious, despite its inner working is very similar to an LLM

Why?

Because it does not speak.

LLMs do speak like us and therefore we fall into an illusion:

Since they speak like us, they must be like us.

But this inference is wrong.

LLMs are just a series of matrix multiplications.
They are not conscious, and they will never be.

*
Paper in the first reply