↗ 点击上方“Shirley草堂”关注我
从最近的新增关注人数来看，开学了，考研成绩出来 了，大家又开始发力了。新来的小伙伴们，公众号每周一、三、五推送口译素材，二、四、六推送双语外刊。往期推送的干货都在每天最后一篇推文中。
双语外刊
Intelligence services
智能服务
Investors are going nuts for ChatGPT-ish artificial intelligence ChatGPT式AI让投资者疯狂
Even Elon Musk wants his own AI chatbot
就连马斯克也想造一台自己的AI聊天机器人
SINCE CHATGPT’S launch in November, a mini-industry has defied the broader slump in tech. Not a week goes by without someone unveiling a “generative” artificial intelligence (AI) based on “foundation” models—the vast and complex algorithms that give ChatGPT and other AIs like it their wits. On February 24th Meta, Facebook’s parent company, released a model called LLaMA. Elon Musk, boss of Tesla and Twitter, reportedly wants to create an AI that would be less “woke” than ChatGPT. One catalogue, maintained by Ben Tossell, a British entrepreneur, has just grown to include, among others, Isaac Editor (which helps students write essays) and Ask Seneca (which answers questions based on the stoic philosopher’s writings). ChatGPT may be much talked about and, with over 100m users, talked to. Yet Mr Tossell’s database hints that the real action in generative AI is in all manner of less chatty services enabled by foundation models.
自ChatGPT于去年11月推出以来，有一个迷你产业逆科技界整体的低迷运势而上。每周都有公司推出基于“基础”模型的“生成式”AI，这些模型就是赋予了ChatGPT和其他类似的AI系统以智慧的庞大而复杂的算法。2月24日Facebook的母公司Meta发布了一个名为LLaMA的模型。据报道，特斯拉和推特的老板马斯克希望创建一种不像ChatGPT那么“觉醒主义”的人工智能。英国企业家本·托塞尔（Ben Tossell）维护的一个目录刚刚扩容，纳入了 Isaac Editor（帮助学生撰写论文）和Ask Seneca（根据斯多葛学派哲学家的著作回答问题）等。ChatGPT可能引起了广泛热议，并且有超过1亿用户与之对话。但托塞尔的数据库显示游戏开发中的人工智能 拜拜，生成式AI的真正进展发生在通过基础模型实现的各式各样重点不在聊天的服务上。
Each model is trained on reams of text, images, sound files or other data. This allows them to interpret instructions in natural language and respond with text, art or music. Though such systems have been around for some time, it took a consumer-facing service such as ChatGPT to capture the world’s—and investors’—imagination. As Mike Volpi of Index Ventures, a venture-capital (VC) firm, says, this happened just as his fellow tech backers, burned by the cryptocurrency crash and the empty metaverse, were on the lookout for the next big thing. In addition, even more than web browsers and smartphones, foundation models make it easy to build new services and applications on top of them. “You can open your laptop, get an account and start interacting with the model,” says Steve Loughlin of Accel, another VC firm.
每个模型都要经过大量文本、图像、声音文件或其他数据的训练，让它们能够解读用自然语言写就的指令，并用文本、艺术或音乐的形式做出回应。尽管此类系统已经存在了一段时间，但真正抓住全世界和投资者想象力的却是ChatGPT等面向消费者的服务。风投公司Index Ventures的麦克·沃尔皮（Mike Volpi）说游戏开发中的人工智能 拜拜游戏开发素材，自己和其他高科技风投者在加密币崩盘和空空如也的元宇宙上烧掉了大把的钱，正急于寻找下一个投资的风口浪尖。此外，与网络浏览器和智能手机相比，在基础模型之上构建新服务和应用程序还要更容易。另一家风投公司Accel的史蒂夫·拉夫林（Steve Loughlin）说：“只要打开笔记本电脑，创建一个帐户，就能开始与模型交互了。”
Money is flooding into the business. In January it was reported that Microsoft poured $10bn in OpenAI, the startup behind ChatGPT, on top of an earlier investment of $1bn. Pete Flint of NfX, another VC firm, now counts more than 500 generative-AI startups. They have so far collectively raised more than $11bn—and that is excluding OpenAI (see chart). Mr Volpi talks of a “Cambrian explosion”.
大量资金正在涌入这个业务领域。据报道，今年1月，微软在早前投资了10亿美元的基础上，又向开发出ChatGPT的创业公司OpenAI投资了100亿美元。另一家风投公司NfX的皮特·弗林特（Pete Flint）估计现在有500多家生成式AI创业公司。到目前为止它们总共融资超过110亿美元，这还不包括OpenAI筹到的钱（见图表）。沃尔皮提到了“寒武纪大爆发”。
So which generative-AI platforms will make the big bucks? For now, this is the subject of head-scratching in tech circles. “It’s just not clear if there will be a long-term, winner-take-all dynamic in generative AI,” wrote Martin Casado and colleagues at Andreessen Horowitz, one more VC firm, in a recent blog post. Many startups offer me-too ideas, often more feature than product. Even the resource-intensive foundation models may end up as a low-margin commodity: though proprietary ones such as OpenAI’s GPT-3.5 are ahead, open-source alternatives aren’t far behind.
那么哪些生成式AI平台能够赚到大钱呢？目前这是科技圈子里令人挠头的话题。“在生成式AI里是否会出现长期的赢家通吃的生态，现在还不清楚。”另一家风投公司Andreessen Horowitz的马丁·卡萨多（Martin Casado）及其同事最近在一篇博客文章中写道。许多创业公司提供的都是跟风的理念，通常只是添加功能而不是推出新产品。即使是资源密集型的基础模型最终也可能变成一种低利润的日常商品，因为尽管OpenAI的GPT-3.5等专有模型目前在领跑，但开源的替代方案也并没有落后很多。
Generative AI is also tiptoeing into a legal minefield. The models often get things wrong. And they can go off the rails. Sydney, the chatbot Microsoft is developing for its Bing search engine using OpenAI’s tech, has insulted a few users and professed its love to at least one (it has since been reined in). AI platforms may not enjoy the legal protection from liability that shields social media. Copyright holders of web-based content on which existing models are being trained without asking permission or paying compensation are up in arms. Getty Images, a repository of photographs, and individual artists have filed lawsuits against AI art-generators such as Stable Diffusion. Stable Diffusion says, “We take these matters seriously. We are reviewing the documents and will respond accordingly.” News outlets fear text-gobbling AIs, too.
生成式AI也在试探法律雷区。模型经常会出错。而且它们可能会突然行为失常。Sydney是微软正在用OpenAI的技术为其搜索引擎必应开发的聊天机器人，它已经侮辱了几名用户，并向至少一名用户表达了爱意（此后已受规束）。AI平台可能享受不到庇护了社交媒体的免责条款的法律保护。一些现有模型在未经许可或支付费用的情况下使用基于网络的内容进行训练，这些内容的版权持有者已经开始强烈抗议。照片库Getty Images和艺术家个人已对Stable Diffusion等AI艺术生成器提起诉讼。Stable Diffusion表示：“我们会严肃对待这些问题3D场景，现在正在查看相关文件，将作出相应的回应。”新闻媒体同样对这类大量使用文本的AI心存忧虑。
OpenAI is already downplaying the launch later this year of GPT-4, the highly anticipated update to its foundation model. It won’t temper VC types’ appetite for generative AI. For more risk-averse investors, the safest bet at the moment is on the providers of the ample processing power needed to train and run foundation models. The share price of Nvidia, which designs chips useful for AI applications, is up by 60% so far this year. Cloud-computing services and data-centre landlords are rubbing their hands, too. Whichever AI platform comes out top, you can’t go wrong selling picks and shovels in a gold rush. ■
OpenAI已经在淡化今年晚些时候将推出的GPT-4，这是其基础模型的升级版，备受期待。这不会抑制风投公司对生成式AI的兴趣。而对于更厌恶风险的投资者而言，目前最安全的赌注是那些提供训练和运行基础模型所需的充足处理能力的供应商。今年迄今为止，设计AI应用芯片的英伟达（Nvidia）的股价已经上涨了60%。云计算服务和数据中心业主也在摩拳擦掌。无论哪个AI平台脱颖而出，在淘金热中镐和铲子总归是不愁卖的。
