发表时间：2024.12.20     作者：Jingyi     来源：Bloomberg     阅读：41

封面故事：Before personalized photo cards, there was Hallmark. Photographer: Jim Watson /Getty Images

原标题：AI 管家来到你家

翻译者 Jingyi，百度翻译助攻。

01 我们已经有 AI 管家了，您呢？

圣诞节还剩下不到 1 周，我们家的贺卡已经在邮箱了，部分感谢 AI。对于我而言，将这些季度的问候放一块儿，耳罩，或者令人讨厌的涂涂画画。今年，我非常惊讶，AI 干这些活的时候，是多么轻松，即便感觉上像在作弊一样。

这是大问题，关于因以下事情而生气，即难以描述的好品质，关于人类的创造性，一场争论无可避免地变得更加复杂，当爱和大方以及其他的讨厌的 IRL 情绪被纳入。这似乎错了，几乎是全自动的能力，由 Apple 智能、或者必应的分包 Crosby-esque、Google Gemini chatbot。是不是？但是我仍然不觉得内疚，我们假期的贺卡有点 AI+，即便有人会批判那些橱柜里的精灵。

earmuffs 耳罩
je ne sais quoi 难以描述的好品质。法语。
sprinkling 少量的

有个问题爆发了，关于 AI 是否拉扯人们的注意，使得人们无法做出良好的作品，就是争论关于无法避免变得复杂这件事，特别是当爱情、慷慨，以及恼人的现实生活中的情绪出现的时候。这似乎错了，全自动的礼物，例如苹果智能产品，或者分包的必应 Crosby-esque （bing 语言类产品），或者谷歌的 Gemini chatbot。是不是？然而，我却未感到内疚，在我们假期贺卡上，有些少量的 AI作品，甚至在充满判断的见证之下，就像橱柜上的精灵那样捉摸不透。

pesky 口语，annoying
IRL in real life IRL —— 是不是觉得有共鸣？有共鸣就对了！
sprinkling 少量

大麻烦是，总是发现家庭照片讨论到一半就没结果。但是，随着谷歌、Adobe 的 AI 编辑开始工作，他们开发的 Express 或者 Lightroom APP 上线，清理拼贴小菜一碟，这些拼贴画包括我、我夫人、小孩、小狗 Derby 都是一样的。只要点击几下鼠标就能删除不想要的元素，或者将多云天气改为晴天天气。我们中的一个人的形象，用于忙碌的商业区形象，谷歌的“魔法橡皮”代替了一系列过路人以及背景里的停车，用一些安静的石头替换，以及空的砖瓦和人行道。

hassle 麻烦
halfway presentable 一半也是能接受的。译为，半途而废。
cinch 小菜 这里的含义，a piece of cake
collage 拼贴画
busy downtown setting 忙碌的市中心设定
take a few taps to 点击几下鼠标就干什么

这使得一些形象产生误导了吗？是的。但是这也只是另一个更快的版本，关于长期以来人们在创造什么，关于专业的摄影方面的创造，Photoshop、Instagram、滤镜或者类似的。这是一些小的尝试，从设计业务的角度来说，例如 Shutterfly 或者 Zazzle。“生成式的 AI 其实是一种工具，给到客户杠杆。给到人们 10 - 100 倍的加速，而之前是完全不可能的。”Andrew Laffoon 说道。Mixbook 合著者，一位非常著名的数字服务商，他们专门制作假期贺卡，以及照片册。

Mixbook 已然出台一系列 AI 相关产品。有些 AI 产品是说明文字生产者，例如，帮助用户化进行头脑风暴，基于一些照片主题。针对我女儿的照片，一个小女孩正在偷狗粮的照片，这暗示了一系列轻松的标语（“友情来自于面包”），同样，一些短诗也是为个性化场景而改变。Laffoon 将以下这些事情可视化了，即能够和 AI 对话并要求，精确的需求。“你可以告诉 AI，将这些事情变得更加感性一些、或者更有趣、更精简、更冗长。”他说到。“让我们诚实一些，并不是所有人都有资格去写一段祝福语作为标志。”

caption 说明文字
dog treat 狗粮
a slew of 大量
lighthearted taglines 轻松的标语
envisions 想象，将什么可视化。

用非常滑稽的方式，AI 某种程度是完美的关于标志式的语句的继承者，这些都是匿名的，都是外包的，都是老生常谈的，某种程度也是被社会接受的，尽管他们拥有空虚的氛围。不同于只是获得一系列意见，关于如何在 CVS 或者 Walgreens 写出感性句子的意见，现在你可以获得源源不断的选择，多亏了 AI。对于改照片也是一样的。Laffoon 说到，AI能够提升照片质量，而不是生产他们。“我们并不制作生成式 AI，假装一场去意大利的旅行，或者混入你最喜爱的名人群体。”他说道。“一点点的夸张是OK的，但是，好的故事来源于什么是真实的。”

vacuous vibe 空虚的氛围
vibe 气氛，氛围

Laffoon 指出，我们已经卸载了这样的工种，从电脑上卸载了十几年了，从他们被事先定义好的模板，从按钮到移除红眼（这些被定义的功能），现代 AI 某种程度代表了新标准的创新、情感的动画。这可能可以解释，为什么现在的品牌，听起来人们会担忧以下这些事，即过多机械化地改变其生产过程。在邮件中，一位发言人，来自 Shitterfly，说到 AI 是策动“人们链接”，而非代替了你的朋友，同时， Zazzle 代表说到，AI 只是放大了一些方式，“保持了真实性，这些真实性使得每一件创作都独一无二。”

这是件非常有趣的事情，见证这些公司是否能够达到合适的平衡，在 AI 以及真实之间。可能，AI 作品将会变得太平淡，或者太造作。可能，假期卡片是足够浅显的语言需求，收到卡片的人，并没有必要去分辨其中差异。如果一部电影，暗示太多电影本身关于未来的猜想，拥有人类生产的语句，是有更多价值的。即便是电影世界中的高级 AI，Joaquin Phoenix 主演了一场失败的浪漫，那些全职的人，正在为客户手写情书，暗示了人类情感，仍然是非常重要的，即便我们难以说服自己。

generic or phony 通用的、普通的，要么就是虚伪的，虚假的。这句话的意思是，这种语言生成，要么平淡无奇，要么就是听起来很造作，和人类写作还会有差距。
penning handwritten love letters 手写情书

“表达情感，对大多数人来讲，是非常困难的。既混乱又对环境敏感。”Laffoon 说到，“当你拥有一些人帮你时，就像标准贺卡一样，那些帮助给到你能力去说些什么，可能你自己是不会说的。”

02 高级 AI 变得比预期更加慢了

2 年来，OpenAI 几乎主导了生成式 AI 的时代，引入了 ChatGPT，向人们展示了模糊的图景，即某种方式上的技术胜人一筹。OpenAI 以及他的主要竞争者，Anthropic，谷歌，Meta，已然出版了一阵掀动技术前沿的 AI 模型，每个版本都比之前更好。这现在是硅谷的福音，更多计算力，更多数据，更大的模型，导致了基础的变化，在 AI 领域。这些技术将会整个转移到工业，在之后的几年。

upmanship 胜人一筹
a flurry of 一阵
gospel 福音

然而，关于发展速度的威胁，开始显现，在第二个 ChatGPT 到来之前。2024 年，OpenAI 以及两个其他 AI 领衔公司，正好踩到了绊脚石。在 OpenAI 以及谷歌，一些软件没能达到内部预期，与此同时，Anthropic 长期等待的模型的时间表，这家公司由 OpenAI 前员工创立的。公司宣布成立之后，创始人就从公司离开了（slipped after it had already been announced）。如果生成式的 AI 的进步，在能容忍情况下变慢的话，这会带来问题，技术是否能够保证这样一件事，即行业内的顶尖人才，已然为 AI 贡献了。识别各种方式，推动 AI 发展至下一个阶段，将会是该领域 2025 年基础的挑战。

stumbling 绊倒
stumbling block 绊脚石
slipped after it had already been announced 宣布之后就溜走了

在一个以自己的创新为傲的行业，Rachel Metz 写到说，公司正在找不同的方式，推动更多模型发展：AI 巨头寻找新的策略，尽管“可预见的结果很少”，这样的情况不复存在了。

03 Snapchat 在青少年芬太尼危机中欠了什么？

“我是个幸存者，但可能对你，Snapchat CEO 来说，不是好事情。”Michael Brewer 不能完成他的话。青少年的演讲，很慢而且被人诋毁，他被打断了，由于其不自觉地呕吐反射，在喉咙与舌头之间。这是一种大脑损伤的症状，由芬太尼引起。

slurred 诋毁
gag reflex呕吐反射咽反射；呕反射
involuntary 非自愿的

这是一个闷热的 8 月下午，在 Jacksonville，Florida。Michael 的腿被包扎了，因为被卷入一个练习自行车内，在一家康复中心，这个地方他学着重新获得他身体的控制力，以及他希望，能再次走路。Eminem 通过耳机输出（pumps），雨水通过窗户倾泻入内，Michael 也看不到，因为芬太尼也使得他致盲。他吞咽、呼吸，以及再次尝试：“我幸存了，但是对你来说的坏消息，snapchat。因为我正在录下我说的话。”

muggy 闷热的
strapped 包扎。缺钱的，手头紧的。
rehabilitation facility 康复设施
pumps 输出
pelts 倾泻

Michael 并非唯一对彭博这么说的人。他也是明星见证人，涉及一场官司，关于 64 个家庭，追逐着 Snap Inc. 的家庭，声称他们公司的 Snapchat APP 助力点燃一场流行病，就是青少年摄入过多药物。Michael 13 岁的时候就联系了交易商（药物交易商），在 Snapchat 上认识的，买了一些，他是很无知的，其实他买的药物和芬太尼有关。他拿到了，眼前一阵黑，呼吸停止。现在 Michael 是 17 岁，他成了明星证人，因为他是两位涉案青少年之一，这两位青少年能描述发生了什么。其他的小孩都死了。

be laced with 给……系上带子

这是 Olivia Carville 新特征的开始，关于小孩的安全，在数字世界。保持阅读。

01 Season's Greetings, From our AI-modified Family to Yours

With Christmas less than a week away, my family’s holiday card is finally in the mail, thanks in part to artificial intelligence. For me, putting together these seasonal greetings normally becomes—earmuffs, Mom—an annoying, last-minute scramble. So this year I was surprised at how easy AI made the process, even if it felt like cheating.

There’s a big question raging about whether AI contributes to or detracts from the je ne sais quoi of human creativity, a debate that inevitably gets more complicated when love and generosity and other pesky IRL emotions get involved. It seems wrong to, say, fully automate gift giving with Apple Intelligence or subcontract Bing Crosby-esque tenderness to Google’s Gemini chatbot, right? Yet I felt zero guilt sprinkling AI on our holiday card, even under the judgmental eyes of the Elf on the Shelf.

The big hassle is usually finding family photos that are halfway presentable. But with all the AI editors available in Google Photos and Adobe Inc.’s Express and Lightroom apps, it was a cinch cleaning up a collage of me, my wife, our baby daughter and our dog, Derby. It took just a few taps to eliminate unwanted elements and alter cloudy lighting to sunny skies. In one portrait of us taken in a busy downtown setting, Google’s “magic eraser” replaced a bunch of passersby and parked cars in the background with quieter storefronts and empty brick sidewalks.

Does this make the image misleading? Sure. But it’s also just a faster version of what people have long been creating with professional photographers, Photoshop, Instagram filters and the like. And it’s a small taste of what’s on the horizon from design businesses such as Shutterfly and Zazzle. “Generative AI is a tool that gives customers leverage—it gives them 10x or 100x more than what was possible before,” says Andrew Laffoon, co-founder of Mixbook, a popular digital service for crafting holiday prints and photo books.

Mixbook has already rolled out a bunch of AI features. Its caption generator, for example, helps users brainstorm text based on photo subjects. For a picture of my daughter playfully stealing a dog treat from Derby, it suggested a slew of lighthearted taglines (“friendship grows with every meal”) as well as short poems personalized for the scene. Laffoon envisions being able to talk with the AI to request exactly what you want. “You can say to the AI: Make it more sentimental, funnier, shorter, longer,” he says. “Let’s be honest, not everyone qualifies to write cards for Hallmark.”

In a funny way, AI is kind of a perfect successor to Hallmark-like messages, which are synonymous with outsourced cliché and somehow societally acceptable despite their vacuous vibe. Instead of getting just dozens of options for sentimental messages at a CVS or Walgreens, you now can get virtually infinite alternatives thanks to AI. The same goes for photo modifications. Laffoon says AI will help enhance pictures, not manufacture them. “We’re not doing generative AI to make up a fake trip to Italy or insert your favorite celebrity,” he says. “A little exaggeration is probably OK, but the good stories are rooted in what’s true.”

Laffoon points out that we’ve offloaded this sort of work to computers for decades, from predefined layout templates to buttons for removing red eye, but modern AI certainly represents a new level of creative and emotional automation. It perhaps explains why brands in the space sound wary of mechanizing too much of the process. In emailed statements, a Shutterfly spokesperson says AI is an enabler of “human connection” rather than a replacement for it, while a Zazzle rep says AI only augments its approach “to preserve the authenticity that makes each creation unique.”

It’ll be interesting to see whether these companies can strike the right balance between AI and authenticity. Maybe AI creations will come across as too generic or phony. Maybe holiday cards are already superficial enough that recipients won’t be able to tell the difference. If the movie Her offers any indication of the future, messages with a human touch will always have more value. Even in the film’s world of advanced AI, Joaquin Phoenix played a lost romantic whose full-time job was penning handwritten love letters for customers, suggesting human sentiment will remain important, even if we have trouble conveying it ourselves.

“Expressing emotion is really hard for most people to do—it’s messy and vulnerable,” Laffoon says. “When you have someone help you, like a Hallmark card, that gives you the ability to say something you might not have said on your own.”

02 Advanced AI Moves Slower than Expected

The two years since OpenAI supercharged the generative AI era with the introduction of ChatGPT have passed in a blur of technological one-upmanship. OpenAI and its primary competitors, Anthropic, Google and Meta, have released a flurry of cutting-edge artificial intelligence models, each more skillful than the last. It’s now Silicon Valley gospel that more computing power, more data and larger models will lead to such fundamental improvements in AI that the technology will transform entire industries within the next few years.

And yet, threats to the pace of development began emerging even before ChatGPT’s second birthday. In 2024, OpenAI and two other leading AI companies hit stumbling blocks. At OpenAI and Google, some software failed to live up to internal expectations, while the timetable of a long-awaited model from Anthropic, a competitor built by former OpenAI employees, slipped after it had already been announced. If progress in generative AI slows in some durable way, it will bring into question whether the technology can ever achieve the more expansive promises the industry’s top innovators have made for it. Identifying ways to propel the AI boom into its next stage will be the field’s primary challenge in 2025.

In an industry that prides itself on innovation, Rachel Metz writes, companies are looking for different ways to push models forward: AI Giants Seek New Tactics Now That ‘Low-Hanging Fruit’ Is Gone.

03 What Snapchat Owes in Teen Fentanyl Crisis

“I’m a survivor, and that’s bad for you, CEO of Snapchat, because, uh, uh, uh …” Michael Brewer can’t finish his sentence. The teenager’s speech is slow and slurred, interrupted by an involuntary gag reflex as his tongue slides down his throat—a symptom of a brain injury caused by fentanyl poisoning.

It’s a muggy August afternoon in Jacksonville, Florida. Michael’s legs are strapped into a stationary exercise bike at a rehabilitation facility where he’s learning to regain control of his body and, he hopes, to walk again. Eminem pumps through his headphones. Rain pelts the windows, but Michael can’t see it, because the fentanyl also left him blind. He swallows, breathes and tries again: “I survived, and that’s bad for you, Snapchat, because I’m talking on the record.”

Michael isn’t referring only to his interview with a Bloomberg Businessweek reporter. He’s also a star witness in a lawsuit 64 families are pursuing against Snap Inc., alleging that the company’s Snapchat app helped fuel an epidemic of teen overdoses. At 13 he connected with a dealer he met on Snapchat and bought a pill that, unbeknownst to him, was laced with fentanyl. He took it, blacked out and stopped breathing. Now 17, Michael is a star witness because he’s one of only two teens in the case who can describe what happened firsthand. All the other kids are dead.

That’s the start of Olivia Carville’s new feature about child safety in the digital world. Keep reading: Fentanyl Almost Killed Michael Brewer. Now He Wants Snap to Pay.

长按/扫码，有您的支持，我们会更加努力！