Principle of AI #1
"AI systems can do tasks, not jobs"
HB
by Harry Berg
"The current [AI] systems are actually not very good at all at doing whole jobs, they're very good at doing tasks"
Sam Altman - CEO, OpenAI
This is what your job might look like today:
This is what it will look like once you're AI-Enabled:
We're going to go through the AI Toolkit that can be applied to have your specific tasks done for you in the next doc once we're finished with Principle #1 here.
How will AI change your job?
Your job before: Do the task
Your job after: Manage AI systems that do the tasks for you
What do you mean "manage AI systems"?
Manage:
- Set objectives
- Review output
Just like your manager (or you, if you're already a manger) already does, except now you're managing AI systems like GPTs and other AI systems - Read on to see some examples.
I know what you're thinking…
"These AI systems never produce anything as good as I would"
How often have you been let down by the results from ChatGPT, or the draft produced from an AI system prompt box in any of your favourite apps?
But hold on…
- Much of why AI systems don't do what we want is not because they are dumb, it's because we have not asked for exactly what we want.
Most people write terrible under-specified prompts, like these:
1
"Write me a marketing plan"
2
"Summarise these notes"
3
"Create an image of a smiling customer"
The result of not knowing how to control AI systems is this:
If you gave the same 1-sentence briefing to a generalist team member of yours, would they produce anything good either? And how long would it take or how much would it cost?
Stick around and we'll go through a checklist and pro-tips for how to write much better prompts together in another doc soon.
Learning to write improved prompts will help you to get much better results.
2. Today's AI systems are only going to get better
The current AI systems are the worst they ever will be, and will likely get much better quickly.
You can think of today's AI systems as very junior interns that need a lot of supervision, but can do well-defined tasks, and take on feedback.
Examples of how AI systems can be applied to do well-defined tasks
- Use ChatGPT to draft your applications/RFPs/contracts
- Create a custom GPT to draft emails in advance of you reading them, as soon as they arrive
- Create an internal chat oracle to answer questions about your knowledge base and avoid having to trawl through documents to find answers
- Use OpenAI's code interpreter/data analysis feature to analyse data in place of hiring an analyst
- Use GitHub Copilot to write code for you 30% faster
Don't worry if you're unfamiliar with any of these tools, this is just a very light set of examples to help make the info shared in this doc more complete.
Shortly, we're going to go over a comprehensive toolkit.
Additionally, I'll send you a curated version based on the tasks you actually do, if you fill in the worksheet at the end of this doc.
What's next?
Now that you understand AI Principle #1: "AI systems can do tasks, not jobs", it's time to put that into practice.
Fill out this worksheet to let me know which tasks your job consists of. I'll use this information to provide you with relevant case studies and playbooks for how you can use certain tools to help with these tasks.
After that, we'll go through the AI toolkit that you can apply to each of these tasks to have AI pick up some of the work for you.