The AI Task Hierarchy
Avoid wasting time on tasks that AI won't be good at; Custom instructions worth taking a closer look at
Welcome to another edition of the best damn newsletter in AI. Here we’ll break down AI topics that matter, open your mind to use cases, and keep you ahead of the curve.
Our #1 goal is to be useful. So please shoot us an email 📩 if you have questions or feedback, and especially if you implement something we share!
Here's what we're covering today:
How to pick the right things to delegate to AI
Uniquely robust custom instructions for ChatGPT
A check in on corporate AI adoption
Let’s get to it! 👇
The AI Task Hierarchy
AI isn't magic (yet).
Today, we'll talk about a simple guide to know what AI can do well and what it might struggle with.
Meet The AI Task Hierarchy.
We are all about delegating our work to AI here at The AI Exchange.
And in doing that, we noticed a pattern between projects that have been easy to delegate to AI, and those that haven’t.
In short: Tasks higher on the hierarchy are harder for AI. Take more time to get right. And oftentimes need a human to add the “excellence”.
Let's break it down.
How the AI Task Hierarchy works:
First, we divide all work into three buckets:
1. Correct work: These have clear right or wrong answers. Think of it like a true/false question. For instance, finding main ideas in a blog or sorting customer emails.
2. Good enough work: These don’t have to be perfect, just decent. Like writing a simple email or making a summary of a meeting. They should be clear and make sense, but they don't need to be amazing.
3. Excellent work: These are challenging. They need a lot of thought, creativity, and knowledge. AI can do them, but it needs a lot of help and time.
Using the framework
Use this framework to help you decide what projects to begin delegating to AI. If something is "excellent work", try to make it into a task that is "good enough work" or "correct work" so that it’s easier to delegate to AI.
If you can't change an "excellent work" task, it's okay to still use AI. Just think of AI as a helper. You’ll certainly need a person to check the AI's work before it's finished (human in the loop).
Custom Instructions worth using
If you’re a ChatGPT Plus subscriber, you have access to Custom Instructions - which is basically a way to have ChatGPT use the same preamble each time you start a new thread.
People have used these to give context on their business. We’ve used it to preserve important context in extremely long conversations.
But Ethan Mollick had his own use for it: start each conversation off like you are having a chat conversation with a gig worker. It’s certainly one of the more in depth uses of custom instructions we’ve seen, what do you think?
A helpful way to start thinking about using ChatGPT for practical purposes is to genuinely approach it as if it was a messaging app that connected you to a very smart, somewhat naive, slightly alien, well-educated gig worker.
Here is a custom instruction that can be useful, it… twitter.com/i/web/status/1…
— Ethan Mollick (@emollick)
Sep 1, 2023
For your reading list 📚
OpenAI is doing… damage control?…
OpenAI released a set of guides to help teachers use ChatGPT in the classroom
Corporate adoption of AI is about to have a heyday…
Walmart gave their 50K corporate employees access to a Walmart-trained chatbot
And if you’re really nerdy…
Most of ChatGPT’s usage is coming from programming tasks (we’ll do a deeper dive into these study findings soon!)
We'll see you again on Thursday. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected].
🪄 The AI Exchange Team