Part
01
OpenAI升级：像人类一样对话？
Since it launched in late 2022, OpenAI’s ChatGPT has generally fended off suggestions that it has emotions or desires by responding that it’s just an artificial intelligence model. Upgrades announced by OpenAI Monday showed the company apparently trying to make the chatbot act more like a human.
自 2022 年末推出以来，OpenAI 的 ChatGPT 通常通过回应“自己只是一个人工智能”来避免提供有关情感或欲望的建议。然而，OpenAI 近期宣布的产品升级表明，该公司正试图让这款聊天机器人展现出更加人性化的行为。
In demos, the new version of ChatGPT was capable of rapid-fire, natural voice conversations, picked up on emotional cues, and displayed simulated emotional reactions of its own.
在演示中，ChatGPT 的最新版本能够进行流畅自然的对话，识别情感线索，并模拟出自己的情感反应的能力。
在 5 月 20 日的旧金山总部直播中，OpenAI 的首席技术官 Mira Murati 宣布，ChatGPT 将升级并采用一款名为 GPT-4o 的新型、更加强大的人工智能模型。OpenAI 将推出一个新的桌面应用程序，用户可以通过这个应用程序访问使用 GPT-4o 模型驱动的 ChatGPT。此外，现有的移动应用和网页版本也会提供对该模型的支持。不论是免费用户还是付费用户都可以使用这个新的功能。
Murati 表示，与 OpenAI 以前的技术相比，GPT-4o 模型允许 ChatGPT 更快速地响应语音、图像和视频输入。在演示环节中，她和员工一起与 ChatGPT 进行了流畅的对话，ChatGPT 使用一种生动而富有表现力的女性声音回答，能在被打断后灵活地衔接对话。
ChatGPT adopted different emotional tones during the conversation and at times responded as if it were experiencing feelings of its own. When an OpenAI employee said he had been talking about how “useful and amazing” the chatbot is, it responded flirtatiously, gushing “Oh stop it, you’re making me blush.”
在对话中，ChatGPT 展现了多种情感语调，有时，它似乎流露出了自己的感情。当一位 OpenAI 员工称赞这款聊天机器人多么有用时，ChatGPT 以俏皮的口吻回答说：“别这么说了，你都要让我脸红了。"
“This just feels so magical, and that’s wonderful,” Murati said, adding, “over the next few weeks we’ll be rolling out these capabilities to everyone.”
Murati 感叹道：“这感觉太神奇了，真是太棒了。”她还表示，“在接下来的几周内，将把这些功能推广给所有用户。”
Part
02
到底更新了什么？
The first is a powerful multimodal interaction capability.
首先是强大的多模态交互能力。
According to the introduction of the official website, GPT-4o is a new large model for the future human-computer interaction paradigm, with text, speech, image three modes of understanding.
根据官方网站的介绍，GPT-4o 是未来人机交互范式的一种新的大型模式，具有文本、语音、图像的三种理解模式。
用户可以用之前的方式输入文本与之交互，也可以进行语音交流，还能直接播放视频进行互动。
The second area is the GPT-4o's processing power in understanding and generating images.
第二个升级的领域是 GPT-4o 对理解和生成图像的处理能力。
GPT-4o 展示了 17 个场景，包括海报、头像、人物、品牌的实时设计、视频和多人会议的总结无论是设计海报、编辑视频还是总结会议内容，用户只需要通过语音或文本与 GPT-4o 进行交互，无须干预，即可得到高效的结果。
The third place is to have human-like emotions.
第三是类似人的情感。
Take this video, where a newly expectant father looks to ChatGPT-4o for an opinion on a dad joke ("What do you call a giant pile of kittens? A meow-ntain!"). The old ChatGPT4 could easily type out the same responses of "Congrats on the upcoming addition to your family!" and "That's perfectly hilarious. Definitely a top-tier dad joke." But there's much more impact to hearing GPT-4o give that same information in the video, complete with the gentle laughter and rising and falling vocal intonations of a lifelong friend.
以一个视频为例，一位准爸爸想了解 ChatGPT-4o 对一则冷笑话的看法（"你怎么形容一只胖猫？大橘为重”）旧的 ChatGPT4 可能会回答：“祝贺你即将迎来家庭新成员！”或“这太搞笑了。这绝对是一个顶级的冷笑话。”但现在可以听到 GPT-4o 发出笑声，它更像是一个有情感的伙伴，而不仅仅是一个机器人程序。
Or look at this video, where GPT-4o finds itself reacting to images of an adorable white dog. The AI assistant immediately dips into that high-pitched, baby-talk-ish vocal register that will be instantly familiar to anyone who has encountered a cute pet for the first time. It's a convincing demonstration of what xkcd's Randall Munroe famously identified as the "You're a kitty!" effect, and it goes a long way to convincing you that GPT-4o, too, is just like people.
给 GPT-4o 发送一段可爱小白狗的视频，它会立即用“夹子音”回应，像人类对萌宠毫无抵抗力一样。这是兰道尔·门罗(Randall Munroe)著名的“你是一只小猫！ （You're a kitty!）”*效应有力的证明。这样一来，很大程度上，AI跟人并无二致。
译注：“You Are A Kitty Effect” 是一种网络上流行的表达方式，主要用来描述那些在社交媒体或网络社区中因为可爱、无害的形象而受到特别宽容和喜爱的人或事物。
这一更新究竟是好是坏？
.
Is this update good or bad?
PART 03
OpenAI’s tweaks may make ChatGPT more engaging and popular, but they may also change the way users think about the program.
OpenAI 的调整或许能让 ChatGPT 更迷人、更受欢迎，但也可能改变用户对程序的看法。
ChatGPT became a sensation last year thanks to its remarkable ability to answer questions and generate text with what often resembles humanlike understanding. But OpenAI and leading competitors such as Google have so far generally sought to limit the anthropomorphism of their chatbots, by refusing to answer certain types of questions and having models remind users that they are mere computer programs.
去年，ChatGPT 凭借其出色的回答问题能力和类似人类理解的文字生成能力引起了轰动。但 OpenAI 和谷歌等主要竞争者一般都试图限制其聊天机器人拟人化，比如拒绝回答某些类型的问题，并让模型提醒用户：他们只是电脑程序。
Despite that caution, the sometimes lifelike repartee of the latest chatbots can tap into users’ emotions. Some companies have adapted the technology to maximize this phenomenon, offering AI companions that role-play different human emotions or are offered as AI girlfriends and boyfriends. Some AI researchers have warned that the long-term implications of deploying programs that mimic and respond to emotions could be unwelcome.
而最新聊天机器人有时栩栩如生的回答却能触动用户的情感。一些公司已经调整了技术以最大化地利用这一点，如提供能扮演不同人类情感的 AI 伴侣，或作为 AI 女友或男友。一些 AI 研究者则警告，投入使用能够模仿和回应情感的程序可能会带来不好的长期影响。
A team at Google DeepMind last month released a research paper exploring the ethical risks that may arise as AI assistants become more capable. Among other things, the researchers warn that such assistants could become highly persuasive and addictive.
谷歌 DeepMind 团队上月发布了一篇研究，探究了随着 AI 助手能力增强可能会产生的道德风险。除此之外，研究院警告称这些 AI 助手可能会变得极具说服力、令人沉迷。
Last year, a British man pleaded guilty to breaching security at Windsor Castle, saying that a chatbot he had exchanged more than 5,000 messages with had encouraged him to do so.
去年，一名英国男子承认破坏了温莎城堡的安全，并称是一个与他交换了超过 5000 条信息的聊天机器人鼓励他这么做的。
往期回顾 | 少年行刺英国女王 教唆犯竟然是AI？
人工智能仍有很长一段路要走.
There may be a long way to go
PART 04
即使有一个令人信服的语音模式也并非攻无不克，GPT-4o 的一些回答还是过于平淡乏味。
当用户告诉 AI 他们正在准备发布公告时，AI 会回答:“太棒了!公告是一件大事呀，”简直就像世界上最脑残的公关顾问。
And during a Zoom meeting where users discuss the relative merits of dogs versus cats, GPT-4o chimes in with the useless, mealy mouthed response, "I can see the appeal of both," (before impressively summarizing the views of three different human speakers by name).
在一次 Zoom 会议上，用户讨论狗和猫的相对优缺点，GPT-4o插嘴说了一句没用的的话:“我觉得两者各有千秋。”然后它指名道姓地总结了三个不同发言人的观点。
Beyond blandness, we also worry about how an LLM's tendency toward hallucinating incorrect information will mix with this new interaction model.
除了回答平淡无奇之外，我们还担心大语言模型会输出虚构的错误信息，这种行为可能在新的交互模式中更加突出。
It's one thing to see a confidently wrong answer in a text chat, it's another to hear a chatbot give a small chuckle as it tries to gently tell you that 2+2 = 5 or some other such obviously wrong statement.
在短信聊天中看到一个自信满满的错误答案是一回事，听到聊天机器人轻声笑着试图告诉你 2+2=5 或其他明显错误的说法，又是另一回事了。
But even some of GPT-4o's snafus can seem hilariously human in these demos. After the AI assistant unexpectedly breaks into French during a rendition of "Take Me Out to the Ball Game," it owns up to its mistake with a "Sorry guys, I got carried away and started talking in French. Guilty as charged!" The way that GPT-4o can instantly realize its specific error immediately after the fact is baffling and hilarious in equal measure.
但即便是 GPT-4o 的一些小失误，也能体现出人类特有的幽默。当 AI 在演唱《带我出去看球赛》(Take Me Out to the Ball Game)时意外地说了句法语，之后它承认了自己的错误，说:"抱歉大家，我太激动了，开始说法语了，我承认自己的错误！” GPT-4o 表现得像个真人一样，认识到了自己具体的错误。
演示中，当 GPT-4o 照常运行时，也很人性化。
举个例子，它帮助叫洛基（Rocky）的用户为 OpenAI 的面试做准备。它先是笑着开了个自嘲的玩笑，说 OpenAI“听起来有点耳熟......”，然后马上转为支持的语气，祝贺 Rocky 获得了这么好的机会。
When Rocky asks about looking presentable, you can hear GPT-4o instantly shift to its most gentle, diplomatic tone when it says, "You definitely have the 'I've been coding all night' look down, which could work in your favor..."
当洛基问如何才能得体应对时， GPT-4o 立即转变为极其温和、礼貌的语气说：“摆出一种‘我通宵打代码’的疲劳神情，可能会对你有利。”
But I really started to marvel at GPT-4o's ersatz humanity when Rocky put on a goofy-looking hat, and the chatbot literally laughed in his face, calling the topper "quite a statement piece." When GPT-4o tells Rocky, "I mean, you'll definitely stand out, though maybe not in the way you're hoping for an interview," I couldn't tell if the AI assistant was giving the devastating truth in a way only a true friend can or simply acting as a new, digital version of a Middle School Mean Girl.
但当洛基戴上一顶傻乎乎的帽子时，聊天机器人真的当着他的面大笑起来，称这顶帽子“相当有个性”。GPT-4o 告诉洛基：“我的意思是，你肯定会脱颖而出，尽管不是以很体面的方式。”很难判断这到底是一种以好友身份提出直击要害的真实坦率，还是像有些初中生那样恶毒的霸凌语气。
原文链接：
https://www.wired.com/story/openai-gpt-4o-model-gives-chatgpt-a-snappy-flirty-upgrade/
https://arstechnica.com/ai/2024/05/is-the-world-ready-for-an-ai-assistant-that-can-laugh-in-your-face/
编译 | 杨嘉伟 玛迪娜 曹蕾 蒋渝可 王培霖
排版 | 徐城铭
点击关注，携手成长👇