On-boarding your AI Intern
There's a somewhat weird alien who wants to work for free for you. You should probably get started.
Let’s get to work.
In previous posts, I have made the argument that, for a variety of reasons, it is better to think of AI as a person (even though it isn’t) than a piece of software. In fact, perhaps one of the most interesting aspects of our current AI moment is that several billion people just got free interns. They are weird, somewhat alien interns that work infinitely fast and sometimes lie to make you happy, but interns nonetheless.
So, how can you figure out how to best use your intern1? Just like any new worker, you are going to have to learn its strengths and weaknesses; you are going to have to learn to train and work with it; and you are going to have to get a sense of where it is useful and where it is just annoying. The stakes for this are quite high. People using AI have 30-80% higher productivity in some writing and coding tasks, and often feel happier having offloaded their most annoying work. That is a big incentive to learn to work with your intern.
And this is really your intern. What it is worth using it for will be different for everyone, and the value will vary by task. Different people with different preferences may find very different uses. If you are a strong writer with a particular voice, you may never want to use AI for writing. If you are continually paralyzed when faced with a blank page, AI may be more useful. No one can provide clear guidance or a magical prompt; you are going to have to figure it out yourself. Your goal is to learn enough about your AI intern to fill out this chart:
Just Me Tasks are those where the AI is not useful and only gets in the way, at least for now. They might also be tasks that you believe strongly should remain human, with no AI help. As AI improves, the latter category likely becomes more important than the former.
Delegated Tasks are those that you assign the AI, and which you expect to review and offer oversight (remember, the AI makes stuff up all the time), but ultimately do not want to spend a lot of time on. This is usually stuff you really don’t want to do and is low importance, but time consuming. You are going to hope that AI keeps getting better at this in the future.
Centaur Tasks. Gary Kasparov popularized the term centaur for tasks that combine human and AI work, like the half-human/half-horse of myth. Centaur tasks are those that you will learn to integrate AI deeply into your workflow and work. For example, when I am writing something that is more technical than a blog post (these tend to be mostly Just Me Tasks for now, hence the grammar errors), I will throw paragraphs and even sentences to the AI for help, have Bing look up sources and explain them in different ways, and generally move back and forth between multiple AIs and my own work. It stops me from ever losing momentum, and often gives me ideas I never could have come up with before. Centaur tasks are where AI becomes most valuable. Figure out a few of these, if you can.
Automated Tasks are ones you leave completely to the AI, and don’t even check on. Perhaps there is a category of email that you just let AI deal with, for example. This is likely to be a very small category… for now. Today, AI makes too many mistakes to use in an automated fashion. (Though, for example, I often ask it to write Python programs to solve problems, and I don't know Python so I just use whatever code it tells me). You will want to keep an eye on the growing capabilities of AI in the future.
Hiring your intern
You are going to need to decide on your intern’s personality and capabilities by picking the AI models you want to use. I have written in the past about the various foundational models, and have kept that guide fairly updated. As of this moment, though, you should probably plan on using one AI connected to the Internet and, likely, a second, fast AI.
There are no good benchmarks or comparisons, but, for most people, your AI connected to the Internet is going to be Microsoft’s Bing in Creative Mode (the purple screen lets you know it is in creative mode) which is GPT-4, but free and connected to the internet. It is also weird. It has a personality and some other constraints that might make it harder to work with (again, like some people you might know). So you will probably want an offline, less opinionated AI to work with on longer projects or exchanges. There are two good options. You can use GPT-4 (which you can get through ChatGPT Plus, for a fee), which is the most powerful model available and has a fairly calm, neutral personality. Or else you can use Anthropic’s Claude, which is not quite as powerful, but has a longer memory and a remarkably pleasant personality (yes, this sounds weird, but trust me, you will know it when you see it). Google’s Bard is very hit-or-miss, even after its updates, so I would skip it for now, though hopefully that will change. If you have specialized needs, like particular programming or language needs, you will also have to learn more about which models work best for that case. If you create images in any way, you will probably want Midjourney. When Code Interpreter is widely released as part of ChatGPT Plus, I would strongly suggest getting it if you work with data in any way.
You also need to figure out your intern’s job description. There are two ways to get started on this. One way is just to ask it what it can do for you directly. But, looking at the replies in the image below, you will notice a few problems: the answers may be naive, or contain factual errors, or may not highlight the most helpful things AI can do. So, while you should try this approach (you AI intern will do it for free), you probably should not trust the intern to know its own tasks.
Instead, you should tell it who it is and what it can do. This might change from task to task, but giving the AI context and constraints makes a big difference. So you are going to tell it, at the start of any conversation, who it is: You are an expert at generating quizzes for 8th graders, you are a marketing writer who writes engaging copy for social media campaigns, you are a strategy consultant. This will not magically make the AI an expert in these these things, but it will tell it the role it should be playing and the context and audience to engage with. It will be you that will need to judge how good it is at these tasks.
Working with your intern
My advice is to include your AI intern in every aspect of your job to see what happens. You need to do this while paying attention to corporate policy, ethics, and privacy concerns (though ChatGPT now has a privacy mode that promises not to use any of the information you enter), but for tasks where it is possible, you should just ask the AI to help and see what happens. Don’t worry about crafting the perfect prompt. Give it messages to respond to, emails to write, ask it to draft reports, have it write code to solve your problems, analyze and summarize data - just do everything.
The initial results are likely to be disappointing. In classes I have taught, students first use AI in ways it is not particularly good at. They ask ChatGPT questions about specific details, yet ChatGPT is not connected to the internet (unless you are using the browsing add-in, which doesn’t work that well), and will often make up answers. They don’t provide enough context in their prompts leading to generic results. They give up after seeing a hallucinated citation, or a made up quote (and it will make stuff up). But remember, just like real people, your AI intern is not an infallible machine, but a weirdly flawed entity. It will have strengths and weaknesses you need to discover, and you will also need to teach and interact with it.
Treat errors from the AI like you would a person. When it makes a mistake, point that out and ask for it to do better. See if it improves, or if you needed to provide additional information in your prompt. Unlike a person, the AI never gets discouraged, so you can ask for volume instead of aiming for a single perfect answer: give me fifty marketing slogans, rewrite this paragraph in ten distinct styles, solve this data analysis problem in as many ways as possible. Sometimes this quantity has a quality all its own.
Like having a real intern, it will be many hours before you really get a handle as to what it can do. I suspect 5-10 hours of actively using AI on work tasks is the minimum. But then, for many people, they start to “get it.” Getting it doesn’t mean that they will use AI all the time. Sometimes they decide that it isn’t that useful for now, and sometimes (based on conversations with a lot of people who want to remain nameless) they use it to do most of their work and don’t tell anyone about it. The more ways you can find for AI to save you time and effort, the more you can benefit in this current odd period, where AI is widely available to individuals, but still mostly not used at the larger corporate and organizational level.
The worst AI intern you will ever have
As you learn about what AI can and can’t do, remember to consider the final column in the chart - the one about the future. The AI you are using is the worst and least capable AI you will ever use. AI is going to be integrated directly in Microsoft Office and Google workspace. Major LLM models will continue to evolve and improve. AI is very likely to get more powerful. So, it is important, as you start to understand AI intuitively, that you think about where AI is evolving. You need to know this for two reasons. First, you want to understand what AI might be able to do in the future to help you become more successful, and plan for that. Second, you want to understand what you don’t want AI to do for you in the future, and plan for that as well.
Either way, your AI interns await: tireless, somewhat alien, but ready to work. What you want to do with this new capability is up to you.
I am going to anthropomorphize AI shamelessly throughout this piece. I know this bothers a lot of people, and I understand why, but it is easier given the context of the post. So just imagine that I put quotes around all of the words that are not technically correct, “The AI ‘knows’ things” rather than “The AI knows things,” etc. I also should note that I am using “intern” in a very loose analogy that offers guidance on how to work with AI. Human interns should obviously not be treated like AI.