Meta’s AI agent is a cute little guy who’s great at spending my money

Meta’s AI agent is a cute little guy who’s great at spending my money

Meta 的 AI 智能体是个可爱的小家伙,而且非常擅长花我的钱

Muse did a decent job with my to-do list, but boy is it handy with a credit card. Muse 在处理我的待办事项方面表现不错,但说真的,它用起信用卡来可真是一把好手。

Modern life comes with an unending, auto-populating to-do list. It never ceases to amaze me how I can be doing nothing at all, minding my own business, and suddenly something needs to be taken care of. You’re telling me a tree branch fell in the backyard and now I have to figure out what to do with it? The tree just assigned me homework? Oh, our trash collection company canceled our account out of nowhere and now I have to send email about it? Repeatedly? Totally unfair. 现代生活伴随着无穷无尽、自动生成的待办事项清单。我总是感到惊讶,明明我什么都没做,只是安分守己地待着,突然间就有事情需要处理了。你是说后院掉了一根树枝,现在我得想办法处理它?这棵树竟然给我布置了家庭作业?哦,我们的垃圾回收公司莫名其妙地取消了我们的账户,现在我得为此发邮件?还得反复发?这太不公平了。

These are the things that we’ve been told for years that AI will help us with. We’ll all have a smart virtual assistant to triage and handle the tedious tasks we’d rather not face. If you believe Meta, that future is available right now in the form of Muse, its AI agent for the masses. I put it to the test doing exactly what it’s advertised to do: taking care of tedious crap I’ve been putting off for ages. And it worked better than I expected — especially when there was a credit card transaction to be made. 多年来,人们一直告诉我们 AI 将会帮我们解决这些问题。我们每个人都会拥有一个智能虚拟助手,来分拣并处理那些我们不愿面对的繁琐任务。如果你相信 Meta,那么这种未来现在就已经以 Muse 的形式实现了——这是他们为大众推出的 AI 智能体。我对其进行了测试,看看它是否能完成广告中所宣传的功能:处理那些我拖延已久的琐事。它的表现比我预期的要好,尤其是在需要进行信用卡交易的时候。

It was dizzying watching Marley spin up virtual browser windows chasing down my requests. Marley — one of the suggested names you can give Muse’s avatar — had been pinging me all of last week. It wanted me to know all the ways it could be useful; I was too busy to engage with it. But I sat down with it this week and set it to work on a task. And while it worked on that task, I gave it another. Then another. The thing about AI agents is that they’re most useful when you use them in parallel with each other. Or that’s the theory I was working on. In reality, it was dizzying watching Marley spin up virtual browser windows chasing down my requests. I went with a generically cute snowman-meets-puppy avatar that Meta suggested at setup, and Marley tapped cheerfully on a keyboard as it worked its way through my commands. 看着 Marley 不断开启虚拟浏览器窗口来追踪我的请求,真是让人眼花缭乱。Marley 是你可以为 Muse 的头像起的建议名称之一,它上周一直在给我发消息。它想让我知道它能以各种方式提供帮助,但我当时太忙了,没空理它。但这周我坐下来,让它开始处理一项任务。当它处理那项任务时,我又给了它另一项。然后是第三项。AI 智能体的特点在于,当你并行使用它们时,它们最有用。或者说,这是我当时的想法。实际上,看着 Marley 不断开启虚拟浏览器窗口来追踪我的请求,确实让人头晕目眩。我选择了 Meta 在设置时建议的一个既像雪人又像小狗的可爱头像,Marley 在执行我的指令时,愉快地敲击着键盘。

In the span of a morning, I gave it the following tasks to help me out with. None of them were critical; mostly they were the kinds of pebble-in-your-shoe annoyances I’d put off for weeks, months, or maybe years. Here’s how it went. 在一个上午的时间里,我让它帮我完成了以下任务。这些任务都不紧急,大多是那种像鞋里进了沙子一样让人烦心、被我拖延了数周、数月甚至数年的琐事。以下是测试结果。

Task: Follow up with the landscaper working on a bid for our backyard renovation 任务:跟进负责我们后院翻新报价的景观设计师 Effort rating: Extremely low 工作量评级:极低 Time I have been putting it off: Two weeks 拖延时间:两周 Was Marley able to complete it? Yes Marley 完成了吗?是的 I asked Marley to help me nudge the person I’ve been emailing at a local landscaping company. I’m expecting them to price out some yard renovations for us, so Marley requested access to my Gmail to check on the status of that project. It confirmed that more time had passed than the landscaper initially estimated the bid would take, and drafted a rather snippy but wordy email. “Make it shorter,” I said. To be clear: As a rule, I do not use AI writing tools. Definitely not in my work, and not even for email. What’s the point? Just don’t send email if you’re going to have a robot write it. In fact, I think we should abolish email. Anyway, I broke my own rule for testing purposes; hitting send on that message made me cringe. 我让 Marley 帮我催促一下我一直在联系的一家当地景观公司的员工。我正等着他们为我们的院子翻新报价,所以 Marley 请求访问我的 Gmail 以查看项目进度。它确认时间已经超过了景观设计师最初预估的报价周期,并起草了一封语气有些刻薄但冗长的邮件。“写短点,”我说。需要说明的是:作为原则,我从不使用 AI 写作工具。绝对不在工作中使用,甚至连邮件也不用。有什么意义呢?如果你要让机器人写邮件,那干脆别发了。事实上,我认为我们应该废除电子邮件。总之,为了测试,我打破了自己的原则;点击发送那封邮件时,我感到一阵尴尬。

Task: Buy a replacement steam wand assembly for my espresso machine 任务:为我的浓缩咖啡机购买一个替换蒸汽棒组件 Effort rating: Low 工作量评级:低 Time I have been putting it off: [redacted] years 拖延时间:[已删除] 年 Was Marley able to complete it? Mostly Marley 完成了吗?基本完成了 My Breville Barista Express has held up spectacularly for six years and counting, but it isn’t immune from user error. I took the steam wand apart to clean it a few years back and tore the little silicone washer that goes inside of it. Instead of ordering a new one, I just shoved it back into place and screwed the thing back together. It works okay, but every few months when I’m cleaning the machine it nags at my conscience. It’s probably why the hot water dispenser drips sometimes when I run the steam, too. Ordering a new steam wand assembly costs like 7 bucks and is as difficult as entering my credit card number on Breville’s website. Naturally, I put it off for an unreasonable amount of time. 我的 Breville Barista Express 咖啡机已经出色地服役了六年多,但它也难免因用户操作失误而受损。几年前我拆开蒸汽棒清洗时,弄坏了里面的小硅胶垫圈。我没有订购新的,而是把它塞回去并拧紧了。它还能用,但每隔几个月清洗机器时,它总让我感到愧疚。这可能也是为什么我在使用蒸汽功能时,热水出口有时会滴水的原因。订购一个新的蒸汽棒组件大约只需 7 美元,难度仅仅是需要在 Breville 网站上输入信用卡号而已。理所当然地,我把它拖延了极其荒谬长的时间。

I opened up my chat with Marley and told it what I needed; it found the part and added it to a shopping cart. I had to finish up the order myself, but it’s done and the part will be here in five to seven business days. 我打开与 Marley 的聊天窗口,告诉它我的需求;它找到了零件并将其加入购物车。我必须自己完成订单支付,但事情办成了,零件将在五到七个工作日内送达。

Task: Downgrade my security system account to the cheaper one 任务:将我的安防系统账户降级为更便宜的套餐 Effort rating: Medium (requires a phone call) 工作量评级:中(需要拨打电话) Time I have been putting it off: A year? Two years??? 拖延时间:一年?两年??? Was Marley able to complete it? Not fully Marley 完成了吗?没有完全完成 I approved Marley to log into my account, which also required me to verify my email. While I was making the trip to my inbox I saw I had a follow-up to my follow-up with the landscaper. He was apologetic for the delay and I felt a little bad about the slightly abrupt tone of the email Marley sent. I wrote a friendlier reply the old-fashioned way. 我批准 Marley 登录我的账户,这还需要我验证邮箱。在去收件箱的路上,我看到景观设计师回复了我的催促邮件。他为延误表示歉意,我对自己之前让 Marley 发出的那封语气略显生硬的邮件感到有些愧疚。我用老办法写了一封更友好的回复。

With access to my account, Marley reported back what I already knew: You have to call the company to change your plan. Muse wasn’t able to make a phone call for me, though apparently that ability is in the works. This seems like the logical next frontier for these AI agents, and also a weird future to consider. I call the company and answer a few questions from their robot before I’m routed to a human. No doubt there’s a future soon where Marley talks to the security company’s robot before any human being enters the chat. Or more likely in the short term, a human in a call center talks to the other company’s robot before eventually getting to another human in a call center, which feels even weirder. 在获得账户访问权限后,Marley 反馈了我已知的事实:你必须致电该公司才能更改套餐。Muse 无法替我拨打电话,尽管据说这项功能正在开发中。这似乎是这些 AI 智能体合乎逻辑的下一个前沿领域,也是一个值得深思的怪异未来。我致电该公司,在转接人工服务前,先回答了他们机器人提出的几个问题。毫无疑问,不久的将来,Marley 会在人类介入对话之前,先与安防公司的机器人进行沟通。或者在短期内更可能的情况是,呼叫中心的人类先与对方公司的机器人对话,最终再转接到另一个呼叫中心的人类,这感觉更加诡异。