In this issue

• Workplace AI apps get huge power boost

• ChatGPT changes how we write at work

• Slow down ChatGPT for better results

• Is ChatGPT really becoming less popular?

• Why we’ve created The AI Writer

LATEST DEVELOPMENTS

Huge power boost for thousands of workplace AI apps

Thousands of apps are about to become more powerful now that millions of developers have access to the latest AI language models.

Something big has just happened in AI, in case you missed it.

News outlets focused last week on whether ChatGPT is on the way out (spoiler alert: it’s probably not – see our other story below). But the company behind it has made a big move that most failed to notice.

1,000+ AI features already

This has already been a huge year for AI workplace apps. Tech companies have added more than 1,000 AI functions to their software this year so far. They claim they can do everything from conducting financial audits to analysing business conferences.

If you’ve tried any AI apps, though, you may have noticed that their performance frequently falls short of expectations. Often, that’s because most are powered by OpenAI’s older language models.

That’s all about to change, because a whole host of apps – even the many free ones – will soon all have access to the tech behind the latest, subscriber-only version of ChatGPT, GPT-4.

And GPT-4 is way more powerful.

For a start, it can handle prompts that are eight times longer, has a better memory and is much better at interpreting instructions written in natural language.

More accurate

OpenAI also claim GPT-4 is 40% more accurate. When testers used it to sit the US Uniform Bar Exam, for example, it performed better than 9 out of 10 human candidates. (Its predecessor had languished in the bottom 10% when it sat it.)

And it can see, so it can process images and tables – a must-have for many business-focused apps.

Early applications include upgrading Be My Eyes, a ‘virtual volunteer’ for visually impaired employees. The free app currently works by connecting the cameras of users’ phones to sighted colleagues who act as their ‘eyes’. But the developer is working on a version that uses GPT-4 to do this instead.

Stripe, the credit card payments processor, is using GPT-4 to analyse customers’ businesses by summarising their websites. It’s also built an app that answers technical questions by scouring all the relevant documentation.

And Morgan Stanley is using GPT-4 to power an internal-facing chatbot that can parse hundreds of thousands of PDFs at once, so financial advisers can use it to answer client questions in real time.

Most apps have been unable to use the new tech until now, however, as OpenAI had restricted its use to only a handful of tech companies. But the company recently announced that it was opening its latest models to the millions of developers on its waitlist.

• In other news, two of the world’s biggest tech giants have also just entered the fray. The Financial Times reported this morning that Meta – Facebook’s parent company – is to give commercial developers full access to its AI code. And yesterday, Elon Musk formally launched his own artificial intelligence company, xAI. The controversial owner of Twitter, SpaceX and Tesla has hired staff from OpenAI, Google’s DeepMind and Microsoft, to challenge ChatGPT’s dominance.

RESEARCH

ChatGPT prompts users to spend more time editing

In experimental writing tasks, one in three ChatGPT users spent twice as much time editing as non-users did. But their efforts didn’t improve their results.

Many professionals are already using AI at work (with or without their employers’ blessing). So are they all outsourcing their monthly progress report to ChatGPT while they sit back, sip their latte and play Angry Birds?

Some maybe. But not everybody.

An experiment at MIT found that, given a choice, two in three workers (68%) did indeed just hand in what ChatGPT wrote for them. But the rest spent most of their time editing rather than writing.

The researchers recruited 444 college-educated professionals who regularly wrote as part of their job. Volunteers included marketers, grant writers, consultants, data analysts, human resource professionals and managers.

Each was given a task that their role would normally include, such as writing press releases, short reports, analysis plans, and delicate emails. Tasks were timed and experienced professionals from the same occupations evaluated the results for writing quality, content quality and originality.

The experiment found that workers who used ChatGPT cut their time to complete their task by over a third (37%). The quality of their work also increased significantly.

More time editing

Volunteers in the control group, who didn’t use ChatGPT, spent around half their time writing the first draft of their assignment. They spent the rest brainstorming and editing.

But the study found that ChatGPT turned that process on its head. Those who used it spent less than half the usual time drafting and twice as much time editing.

It seems they should have saved themselves the bother, though. Evaluators found that the editing generally didn’t improve quality.

In our experience at Emphasis, many workers don’t even get beyond the drafting stage. For them, the writing process is so painful and drawn out that they seldom have the luxury of editing, settling instead for handing in a document that they know they could have made a lot better if only they’d had more time.

We’ve also found that most people don’t know where to start when it comes to editing. Often, they fall back on gut feel and half-remembered lessons from school or changing documents to match those they see most often.

This experiment suggests not just that bots like ChatGPT can improve productivity in writing tasks but that editing will become a must-have skill in the age of AI.

HOW TO

Get quicker results by slowing ChatGPT down

Breaking prompts into simple steps forces an AI bot to pace itself, increasing the chance of a successful response.

Why you need this

There you are, getting perfectly reasonable answers from ChatGPT when suddenly it starts ignoring a key part of your request. ‘Give me six bullet points about x’ you write, and it gives you eight. ‘Brainstorm topics for a report on Y’ you ask, and off it launches into writing the report itself instead.

Why this happens

This happens often because you’re asking it to do too much at once.

When you write a prompt for ChatGPT, you are effectively programming its software to answer your questions (even if it doesn’t feel like it). And as with all programming, complex requests are more likely to go wrong than simpler ones.

What to do

One way around this is to tell the bot exactly how to approach the problem. Break your ‘program’ (the prompt) into several simpler tasks. This will force it to slow down, giving it more time process its response properly.

I would like help to plan a report on [subject]. Its readers will be [audience]. They have [no/some/lots of] existing knowledge and expertise on the subject. 

First, work out what they are likely to know about the subject. 

Then, work out what they are most likely to be interested in. 

Finally, give me a list of five topics I could cover in the report.

TRENDS IN FOCUS

Is ChatGPT’s popularity really dying?

Traffic to ChatGPT’s site saw a big drop in June, but that doesn’t mean the AI bubble is bursting. Seasonality, competing AI apps and even students going on vacation have all had an impact.

It had to happen. After six months of breathless coverage of ChatGPT’s spectacular launch into the mainstream consciousness, a backlash was inevitable.

Sure enough, the past week has seen a rash of news stories reporting that a drop in traffic to the chatbot’s website between May and June heralds the end of the AI trend.

So, are they right?

Well, there were certainly fewer people using it last month. ChatGPT, which had rocketed to an estimated 100 million monthly users in its first two months, recorded a 9.7% drop in traffic in June according to preliminary estimates.

Seasonal changes

If we look at the big picture, though, we see factors at play that have little to do with user satisfaction or fatigue.

First, the drop may have been more down to seasonal changes than a slide in popularity. Traffic to all websites in the US also fell by an average of more than 10% last month. And scores of popular social and search sites recorded marked dips in traffic.

These seasonal trends are not unusual.

There’s no shortage of other theories, either. The Washington Post, for instance, reported that a growing number of companies are now banning certain employees from using the tool at work.

June also marks the end of the academic year in the US. Students are among ChatGPT’s most enthusiastic adopters. Perhaps there are now just far fewer of them using it to do their homework.

Competing apps

Or maybe it’s just that more people are using ChatGPT’s new iPhone app instead. When it launched in May, five million people downloaded it. So that’s bound to have drawn some visitors away from the site itself.

Finally, ChatGPT’s noisy entry into the wider conversation has inspired a raft of similarly themed AI tools, and they’re all now competing for users and eyeballs.

With more AI chatbots launching every day, ChatGPT may well have found itself a little neglected, as users experimented with new tools and platforms. (Not that that will worry its makers: thousands are powered by its tech anyway.)

As with any meteoric launch, at some point the growth curve was always going to flatten. We’ll have to wait until we’ve been through the hype cycle before we really know what’s going on.

– Christian Doherty

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.