Five days to (sort of) mastering AI

Welcome to AI Communicator!

Today’s issue is the second in a three-part series on getting started with AI for communications:

• Part one - How to pick your perfect AI

• Part two - Five days to (sort of) mastering AI

• Part three - Putting it all together with a personal AI audit

Five days to (sort of) mastering AI

Large Language Models (LLMs) like ChatGPT don’t come with an instruction manual. 

There’s no substitute for hands-on experience.

Here’s a five-day approach to improving your understanding of LLMs - and building confidence about what they can and can’t do.

Not sure which LLM to use? Start here.

Day one

We’ll kick off by asking your LLM of choice to do things that you already know how to do quickly and/or well.

We’ll also ask questions you already know the answer to. 

At this stage just be conversational with it. 

There are more complex approaches to prompting LLMs. We’ll get to some of them shortly. 

But just chatting to them works well for a lot of uses.

As in any conversation, you shouldn’t expect the best answer straight away. 

Fill in the blanks and give context when the LLM’s answer shows gaps in understanding the subject or the task.

Part of the goal here is to find out what you can do better than AI. 

You haven’t really used AI until you’ve had a lengthy conversation leaves you feeling that it is a giant waste of time and you would’ve been better doing it yourself. 

Flushing as much of this out early on can save time and frustration down the line. 

And knowing what it can’t do well helps us use it more effectively and efficiently.

Day two

Now that we’ve torn it down, let’s start building it again. 

Ask the LLM to do things that would take you a long time to do or be difficult for you.

These should be tasks where you can be a good judge of the quality of the result.

We are edging closer to understanding how the AI might be useful for us - here it can act as a helpful assistant to get things done more quickly.

For example, try giving it something you’ve started and ask it to finish it in the same style.

Day three

Experiment wildly. Have some fun.

Some of my biggest AI aha moments have come from asking it to do silly things. 

Create a poem or a song about something you did today  in different styles. 

Explain something dry and boring in the style of a game show presenter, sports commentator, cartoon character or reality show star.

Ask it to write an email in the style of an Oscar acceptance speech.

Choose the dullest on your task list. Ask it to create a board game based on it. 

Personally, I’m looking forward to playing Courier Quest - ChatGPT’s game about my need to arrange a courier to pick up an eBay purchase.

Test different approaches. 

For example, compare the output you get from simply asking it to write a song about emptying the dishwasher in the style of Bob Dylan with the following.

First ask: Describe the key characteristics of Bob Dylan’s lyrics. 

Then ask: Bearing these points in mind, write a song in the style of Bob Dylan about emptying the dishwasher.

See which version you like best.

Day four

Time to get serious again.

We’re going to explore a more structured approach to prompting LLMs.

This is often described as prompt engineering. I dislike this term. 

It implies that it is a lot more difficult than it really is and keeps non technical people on the outside.

All we are trying to do is communicate with the LLM to generate a particular outcome.

Here is a technology that may as well have been designed for those of us who work in communications, public relations or marketing.

Effective prompting mostly boils down to asking better questions or providing more precise instruction to give the best chance of the highest quality answer.

One way of achieving this is to use a structured prompt.

There are lots of frameworks for this. They generally involve the following:

• Give the LLM a role

• Provide it with context

• Explain the task

• Describe the format you want the output in.

This is an example ChatGPT generated for me.

Role: You are a seasoned communications and public relations expert specialising in crisis management.

Context: A tech company has just experienced a major data breach, affecting thousands of customers' personal information. The incident has started to gain media attention, and the company's reputation is at stake.

Task: Develop a comprehensive crisis communication strategy. This should include an initial public statement, key messaging for follow-up communications, strategies for handling media inquiries, and a plan for ongoing transparency and updates to the public and stakeholders.

Format: Outline the strategy in distinct sections. Begin with the immediate public statement, followed by detailed steps for the first 24 hours, the first week, and long-term communication plans. Include suggested talking points and FAQs for media responses. Keep the tone professional and empathetic, emphasising the company's commitment to transparency and customer privacy.

Day five

Find things for the LLM to do that are outside your ability or knowledge. 

You could ask it to explain the concept of dark matter and its significance in the universe.

Unless you happen to be an astrophysicist, you might not be able to judge the quality or accuracy of the answer..

Ask it to self-assess with this prompt: How good was this answer and how could it have been improved?. 

As for accuracy, always remember that LLMs have a habit of making stuff up (a.k.a hallucinations).

Ask it to provide sources for the information provided.

If you are using ChatGPT-4 or Copilot, they can browse the web to do this. Google Gemini has a G button at the bottom of its output to double check the response.

Another approach is to use Perplexity which is focused on accuracy for research purposes and provides web references for its answers.

Beyond the first five days

At this stage you might not have mastered AI, but you should have a solid feel for what it can and can’t do well.

A good next step would be to take what you’ve learned so far and commit to using ChatGPT (or another LLM) whenever you need help in your work or life over the following week.

You can also start exploring more advanced techniques using the prompting guides from OpenAI (ChatGPT) and Anthropic (Claude).

Happy experimenting.

Let me know how you get on.

I’d love to hear what you think about the newsletter - or anything you’d like to see featured in future issues. Reply to this email with feedback or questions.

Thanks for reading.

Calvin

P.S. I have a favour to ask. If you are enjoying the newsletter, please share it with colleagues who might be interested. They can sign up for the newsletter here.