In this issue

• Get more done in less time

• Using ChatGPT for book summaries

• How to tell if you’re talking to a bot

• Why we’ve created The AI Writer

NEW SERIES

Me, myself and AI

In the first of this new series, I report on how I used AI to help pull together one of our key stories this week. The techniques could help you keep up with industry news, expand your knowledge quickly and draft examples of key concepts to use in reports.

Anyone trying to use AI now faces a bewilderingly wide choice of options. As well as chatbots like ChatGPT, literally thousands of apps now come with AI functions.

It’s enough to make you run for the hills, hide in a cave and work with nothing more high tech than a spiral-bound notebook and an HB pencil.

That’s probably why one of the most common requests I get is for more practical help on how to use ChatGPT for work.

Reader, I am here to serve.

The best way to approach the AI deluge is to focus not on what apps to try but on what you’re trying to achieve. What’s the problem you’re trying to solve, and what tools can help?

To get you thinking along the right lines, I’m going to start using The AI Writer to chart my own journey with AI and reveal how it’s helping me (or not) in my job.

Let’s begin with a case study that’s almost literally up to the minute: how I used AI to help compile this week’s issue.

I now spend most of my week researching and writing, so I’m open to trying anything that might save time without sacrificing quality.

Arguably, AI is now improving quality, as it has freed up some of my time and brain power to think more about how I can better help my readers. (Hence this new series.)

Here’s how I use it – and how you might want to, too.

Content monitoring

I keep up with the news using the app Feedly to monitor the content of key news websites. (It’s an RSS reader that automatically tracks updates.) Feedly now has an AI function that suggests and summarises content from other sites that I don’t normally check, based on what other content it’s seen me read.

More than once, it’s alerted me to stories that I would have missed otherwise.

Research

I always make sure I go back to original published research before I report on it. But reading complex documents like academic papers can take a lot of time and headspace. To help, I use an AI-assisted app called PDF.ai to read those for me.

This lets me interrogate the document by typing questions in the app’s chat window, as if I were interviewing one of the authors themselves. When the bot answers, it also links to the relevant section so I can find out more and check that what it’s told me is correct.

I used the app to quickly get up to speed on the research for this week’s article, ‘How to detect a customer service chatbot’ by uploading the original research paper and asking questions about it.

Security should never be far from the mind when uploading documents to an AI chatbot or an AI-powered reader. I never upload confidential or proprietary information.

According to PDF.ai, uploaded documents are encrypted before being stored securely. The provider also offers a private document option for processing PDF documents.

The app runs on the large language model behind ChatGPT, which means it connects to servers run by OpenAI (which created that chatbot).

All third-party apps that do that are automatically opted out of allowing OpenAI to train its language models.

That means that any document uploaded to PDF.ai won't be used to train ChatGPT or fed into its knowledge bank.

Documents should also be automatically deleted after 30 days. But, in theory at least, they might still be read by OpenAI’s human checkers before then.

Illustrating complex topics

The above paper also contained its fair share of technical terms. For instance, the researchers write about how bots like ChatGPT are poor at noise filtering. PDF.ai had already told me what that term means, but how could I illustrate it?

To get some ideas, I asked ChatGPT, ‘Give me an example of how LLMs [large language models] struggle with noise filtering.’

Unusually, the answer it gave me was so good that I was able to use it almost word for word. (I rarely do this.) Before I did that, though, I copied and pasted it into an online plagiarism checker to make sure the content was original. (It was.)

WHY TRY AI

101 uses for ChatGPT

Adapt the prompt below and finally cut down that long list of books you’ve been meaning to read but haven’t yet found time for.

It’s estimated that only one in ten people have tried using ChatGPT in their work, even though most say they’ve heard of it. One of the biggest reasons for that is that they don’t know why they should.

So this week, we’re continuing our series on uses for ChatGPT that you might not have considered.

Use # 2: Book summaries

If you’re like most professionals, you’ll probably have a long list of books you keep meaning to read but have never quite got around to. So how about using ChatGPT to summarise some of them?

You can then implement the key points now and decide if you should still read a book or give it a miss.

What to do: Just adapt, copy and paste the text below into ChatGPT.

Summarise [Book title] by [author name], using bullet points. 

Expand each point using sub-bullets.

Give me facts from the book rather than telling me what the book or author do. 

For example, don't write 'The author discusses ...' or 'The book explores'. Instead, summarise the content of that discussion. 

Research

How to detect a customer service chatbot

Not sure if it’s a bot or not? Just try the capital letter test.

As AI becomes more powerful, distinguishing bot from human is getting increasingly difficult. Nowhere is this more true than with customer-service chatbots.

Robotic voices often still sound pretty clunky and artificial (though we’ll be reporting soon on some huge leaps in progress in that area). But with reading, the only voice we hear is in our own heads.

There’s also the problem that our brains didn’t evolve to read and write (as humans have only been doing it for a few thousand years). So reading a chat conversation is harder work than listening to a voice on the end of a phone.

That leaves far less brainpower left over for interpreting whatever we happen to be reading. It also makes it much more difficult to detect whether its author is even human.

Now researchers have come up with a simple test that solves the second of these problems in an instant: insert random words in the text in capital letters.

The test was devised by computer scientists at the University of Santa Barbara and Xi’an Jiaotong University in China. And it exploits a known weakness of the large language models behind bots such as ChatGPT – they struggle to filter out irrelevant information.

Relevant info

Forget live chat for a moment and just imagine that you wanted to use ChatGPT to summarise the following journal entry:

Yesterday, I went to the beach with my friends to enjoy the sunny weather. We played beach volleyball and had a great time. The water was so warm, and the waves were perfect for surfing.

Suddenly, we noticed the sky getting darker, and strong winds started to blow. People around us were getting worried. We decided to leave the beach and head home. Later, we found out that a hurricane hit the coastal city, causing significant damage.

As a human, you could quickly identify the relevant information about the hurricane hitting the coastal city and its impact. You'd focus on the part that mentions the ‘sky getting darker’, ‘strong winds’, people ‘getting worried’ and the hurricane ‘causing significant damage.’

However, an AI bot might struggle with this task because it's trained to generate text based on patterns in its training data. It doesn't have the same understanding of relevance and importance as humans do.

So it might produce a summary like this:

Yesterday, friends enjoyed the sunny weather at the beach, playing volleyball and surfing. Later, a hurricane hit the coastal city, causing damage.

For the same reason, give any literate, adult human this sentence:

IsCURIOSITY waterARCANE wetTURBULENT orILLUSION drySAUNA? 

(is water wet or dry?) and they would quickly decode the question and answer ‘wet’.

But even the latest large language models would struggle to do this (for now, at least).

So, next time you’re in an extended live chat exchange to try to explain your latest gas bill, try scattering random words like SPROCKET and FLANGE throughout one of your questions. If it’s a bot, it won’t be able to answer, whereas a human will.

Be warned though that most human customer-service agents will also probably terminate the conversation immediately. But hey, you can’t have everything.

Why The AI Writer?

When I founded Emphasis, 25 years ago, it was just me, a cat and a kettle. It’s since grown to become the most trusted provider of business writing training in the world and helped more than 80,000 people, from 32 countries. We’ve worked with tech giants, top 10 law firms and major financial institutions, as well as at the highest levels of government. (We’ve even sent trainers to work with clients in the Himalayas.)

I would not claim that we’re experts in everything, as we’re certainly not. But, after working with the authors of around 100,000 documents, it’s safe to say that business writing is something we do know a thing or two about. And that includes witnessing all the various ways in which organisations get it wrong.

Now, we’re at the forefront of using AI to help them get it right.

Of course, I’ve seen nothing like the AI revolution that’s hitting us. Nobody has. But I do feel a responsibility to use our experience to help people navigate this brave new world of communication bots.

So here’s the deal. You get on with your job, while we immerse ourselves in the latest developments in workplace AI. We’ll worry about keeping up so you don’t have to. Then we’ll tell you exactly what you need to know to stay ahead of the curve.