来源：《新科学家》
原文见刊日期：2023年2月25日
Imagine living in a city that, overnight, changed all its road signs. Some of them are still familiar, directing you to the city centre, while others offer more whimsical destinations such as the moon. Some changes streamline the driving experience, while a few are outright dangerous, such as a speed limit of 300 miles per hour. “Oh, we know it doesn’t quite work,” say the city planners. “But the neighbouring city is going to do it and we wanted to get there first. Will you test it for us so we can win?”
想象一下生活在一个一夜之间改变了所有路标的城市。其中一些仍然很熟悉，指引你去市中心，而另一些则提供了更加异想天开的目的地，比如月球。一些改变简化了驾驶体验，而一些改变是完全危险的，比如限速为每小时300英里。“哦，我们知道这不太管用，”城市规划者说。“但是邻近的城市也要这么做，我们想先成为智慧城市。你能帮我们测试一下吗？这样我们就能赢了。”
We would rightly question the competence of anyone wanting to alter vital city infrastructure in this way and yet search engines, a key part of our internet infrastructure, are being subjected to a similar experiment. Google, Microsoft and Baidu are racing to add artificial intelligence chatbots to their search offerings, with the last two attempting to wrest a 90 per cent market share away from Google. But the large language models behind these bots simply aren’t ready for mainstream use.
我们有理由质疑任何人是否有能力以这种方式改变重要的城市基础设施，而作为我们互联网基础设施关键部分的搜索引擎，正在接受类似的实验。谷歌、微软(Microsoft)和百度正竞相在自己的搜索产品中加入人工智能聊天机器人，后两家公司试图从谷歌手中夺取其占有的90%的搜索市场份额。但是这些聊天机器人背后的大型语言模型还没有为主流应用做好准备。
Releasing half-finished software, known as the “minimum viable product” in Silicon Valley circles, allows upstart tech firms to quickly release a new app and iterate based on user feedback, and it has its place. But these chatbots are being created by some of the largest companies in the world, on which billions of people rely, and they have a duty of care. Already, Microsoft has had to rein in the AI version of its Bing search engine, after finding that lengthy questioning can lead to it offering results that appear increasingly bizarre.
发布半成品软件，在硅谷圈子里被称为“最简可用产品”，可以让新兴科技公司快速发布新应用程序，并根据用户反馈进行迭代，这是有其合理性的。但这些聊天机器人是由世界上一些最大的公司创造的，数十亿人依赖这些公司，它们有责任谨慎行事。微软(Microsoft)已经不得不控制其必应(Bing)搜索引擎的人工智能版本，因为它发现，冗长的提问可能会导致它提供的结果看起来越来越奇怪。
Large language models certainly could be transformative in the way their proponents hope, but they have been allowed to escape from the lab far too early. This is partly due to raw economics – the scale of these models makes them too costly for individuals to run, limiting researchers’ abilities to create and study their own – but also to the particular brand of hypercapitalism practised by tech firms, racing to exploit the next big thing. If our chatbot future is indeed coming, there is no need to rush.
大型语言模型当然可以以其支持者所希望的方式实现变革，但它们过早地脱离了实验室。这在一定程度上是简单的经济问题——这些模型的规模使它们对个人来说运行成本过高，限制了研究人员自行创建和研究的能力——但也与科技公司实践的特定超级资本主义有关，它们竞相开发下一个大事物。如果我们的聊天机器人的未来真的来了，真没有必要着急。