Explaining LLMs to my neighbors
John and Geri, my neighbors who are like a second set of grandparents to my son Max, were curious about my new job the other day. Here’s how our conversation went:
Geri: “So Felix, I heard you started working at a new place. What’s your role there?”
Me: “I’ve joined this cool place to help simplify the complexity of business and make it easier for companies and customers to create connections.”
John: “Ah… and how exactly do you do that?”
Me: “Well, we help businesses improve their customer service.”
Geri: “Like making calls to the Comcast cable company less of a headache?”
Me: “Precisely!”
John: “Sounds crucial. But how do you achieve that?”
Me: “I have a few ideas, and I’d love to get your thoughts on them.”
Geri: “Sure, we’d be happy to help.”
Me: “Imagine if Comcast asked us to boost their customer service by finding the best customer representative. We’ve brainstormed a few strategies.”
“Idea one comes from Victoria, who thinks we should start as early as possible. Imagine outfitting newborns with diapers featuring cable patterns, giving toddlers wifi-router-shaped toys, and having first graders ride ethernet-cable bikes. We’d design an entire educational journey from elementary to college, all about learning electronics, the Internet, and any other things that we can imagine that a Comcast expert would need to learn, and maybe even more, just in case, with the sole purpose of crafting the perfect customer service rep for Comcast.”
John: “Hmm….”
Me: “Then there’s Alexander, who agrees that we need specialized retraining to be a successful Comcast CS rep, but he believes Victoria’s approach a bit too much. He suggests we focus on post-college career training specifically for becoming a Comcast customer rep. This would be like an additional four years of graduate study after college, learning advanced knowledge and deep technical information about the Internet, tailored to Comcast’s needs.”
Geri: “I see….”
Me: “Elizabeth has a more grounded approach. She believes that college education is enough and just proposes adding a one-month intensive training for recent graduates, covering everything from tech skills to customer interaction tips.”
John: “That seems more doable.”
Me: “James agrees with Elizabeth but thinks even designing the training is unnecessary. We could outsource it to training specialists.”
Geri: “But isn’t training what your company does?”
Me: “Exactly, which is why Emma suggested simplifying it further. No special training is needed and just give new hires a reference handbook and some cheat sheets. College kids are smart; they’ll just look through the handbook and cheat sheets when customers come.”
Me: “So, which idea do you think is best?”
John: “Alexander’s plan sounds solid for a tech company like Comcast. A general grad might lack the technical know-how. So doing some graduate study is necessary”
Geri: “Yet, a month of training could suffice for customer service basics, like troubleshooting wifi issues, so I like Elizabeth’s idea”
Me: “What about Victoria or Emma’s approaches?”
Geri: “For simple retail jobs mostly about just pointing customers to find where the shelf is for toothpaste, or putting fruits in a basket, Emma’s idea could work.”
John: “Victoria’s plan seems overkill, even for high-tech Internet fields. However I wonder if certain places, like medical, legal, or rocket companies might need it since they are just so special, and the tolerance for risk is so low”
Me: “Right. So basically what I explained are different approaches to developing Large Language Models (LLMs) for customer service. LLMs are ways to train computer computer programs to do something for us. And we can choose to design the LLMs from the most intensive training akin to Victoria’s approach to the minimalist strategy like Emma’s, each with different pros and cons for investments and rewards”
Me: “While Emma’s method is the easiest, it might be too slow for real-time interaction. Think of it as asking a kid to engage in a discussion while the kid is trying to learn how to respond by looking up a dictionary at the same time.”
Me: “Consider public education as basic training for a general-purpose model. It’s crucial for any specialized role later on. The key is the complexity of customer service tasks and the quality of the public education system. In some places, additional training will be necessary.”
Me: “Victoria’s idea, though ambitious, could provide us with a competitive edge by enabling us to offer unparalleled customer service across industries. After all, we are a company that helps others to connect better to their customers, so we should always strive to be the best in doing so. The accumulation of know-how from years of learning from Victoria’s idea sounds crazy and could be particularly special for the business to succeed in the long term. Imagine we might write a whole bunch of parenting best-selling books out of it, and everyone would recognize us as the best in doing this.”
Geri: “It’s all quite intriguing. So, which direction will you take?”
Me: “We’re still exploring. It is all about balancing investment and returns, leveraging what’s available out there (e.g., the public education system), and also anticipating what these would change in coming years. Let’s chat more about it next time.”
邻居John和Geri是我儿子Max的好朋友，他们就像我儿子在美国的爷爷奶奶。前几天，他们对我新工作很好奇。以下是我们的对话内容：
Geri：“所以Felix，我听说你开始在新地方工作了。你在那里做什么呢？”
我：“我加入了这个很酷的地方，帮助简化商业的复杂性，让公司和客户更容易建立联系。”
John：“额……你们是怎么做到的呢？”
我：“嗯，我们帮助企业改善他们的客户服务。”
Geri：“就像让打电话给Comcast有线公司变得不那么头疼？”
我：“正是！”
John：“听起来很关键。但你们是怎么实现的呢？”
我：“我有一些想法，我很想听听你们的看法。”
Geri：“当然，我们很乐意帮忙。”
我：“想象一下，如果Comcast让我们通过寻找最佳客服代表来提升他们的客户服务。我们已经头脑风暴出了一些策略。”
“想法一来自Victoria，她认为我们应该尽早开始。想象一下为新生儿装备上有缆线图案的尿布，给幼儿提供形状像wifi路由器的玩具，让一年级学生骑以以太网线为主题的自行车。我们会设计一条从小学到大学的完整教育旅程，围绕学习电子、互联网以及我们能想象到的任何其他东西，这些都是培养完美的Comcast客服专家所需要学习的，也许甚至更多，其唯一目的是为Comcast打造完美的客户服务代表。”
John：“呃……”
我：“然后是Alexander，他同意我们需要专门的再培训才能成为成功的Comcast客服代表，但他认为Victoria的方法有点过了。他建议我们专注于大学毕业后的职业培训，专门成为Comcast客户代表。这将是在大学后额外四年的研究生学习，学习关于互联网的高级知识和深入技术信息，专门为Comcast的需求量身定做。”
Geri：“好吧😑……”
我：“Elizabeth有一个更接地气的方法。她认为大学教育已经足够，只是提议为应届毕业生添加一个月的密集培训，涵盖从技术技能到客户互动技巧的一切。”
John：“这看起来更可行。”
我：“James同意Elizabeth的观点，但认为连设计培训都是不必要的。我们可以将其外包给培训专家。”
Geri：“但是培训不就是你们公司做的吗？”
我：“对的。我们自己就应该是培训的专家。这就是为什么Emma建议进一步简化。不需要特殊培训，只需给新员工一本参考手册和一些备忘单。大学生很聪明；当客户来的时候，他们只需要查阅手册和备忘单。”
我：“那么，你们认为哪个想法最好？”
John：“对于像Comcast这样的科技公司来说，Alexander的计划听起来很稳固。一个普通的毕业生可能缺乏技术知识。因此，进行一些研究生学习是必要的。”
Geri：“然而，一个月的培训可能就足够了，足以掌握客户服务的基础，比如解决wifi问题，所以我喜欢Elizabeth的想法。”
我：“Victoria或Emma的方法呢？”
Geri：“对于主要是让顾客找到牙膏架子在哪里，或者把水果放在篮子里这样简单的零售工作，Emma的想法可能行得通。”
John：“Victoria的计划即使对于高科技互联网领域来说也显得过分，虽然我想知道对于某些特殊的地方，比如医疗、法律或者火箭公司，可能需要它，因为他们实在是太特殊了，而且犯错误的代价太大。”
我：“对。所以基本上我解释的是开发大型语言模型（LLMs）用于客户服务的不同方法。LLMs是训练计算机程序为我们做事的方式。我们可以选择从最密集的培训（类似于Victoria的方法）到最简化的策略（像Emma的方法），每种都有投资和回报的不同利弊。”
我：“虽然Emma的方法是最简单的，但对于实时互动来说可能太慢了。想象一下，就像让一个孩子一边查字典，一边来跟你参与讨论。”
我：“将公共教育视为一般目的模型的基础培训。很多应用场景，比如说简单的零售店，估计足够了。但是对于任何特定角色来说，这对以后都是至关重要的。关键是客户服务任务的复杂性和公共教育系统的质量。在某些地方，可能需要额外的培训。”
我：“Victoria的想法虽然雄心勃勃，但能够为我们提供在各行各业提供无与伦比的客户服务的竞争优势。毕竟，我们是一家帮助其他人更好地连接他们的客户的公司，所以我们应该始终努力做到最好。从Victoria的想法中积累的知识听起来疯狂，但是这可能给我们打下了坚实基础，和长期成功的优势。想象一下，我们可能会写出一系列畅销的育儿书籍，每个人都会认为我们在这方面做得最好。”
Geri：“这一切都很吸引人。那么，你们将采取哪种方向？”
我：“我们还在决定。这都是关于平衡投资与回报，利用外面现有的东西（例如，公共教育系统），同时也预测这些将来几年会如何变化。下次我们再聊。”
