With everyone on social media now a self-proclaimed expert on A.I. ever since ChatGPT first came to their attention in late 2022, it is only right that I take a closer look. It may feel to you as if we are at peak A.I. saturation right now, but the mania has barely even begun. To term it using the Gartner Hype Cycle, we’re nowhere near the peak of inflated expectations.
Having first explored ChatGPT a few months ago, I expected to find a technology that would make my job easier, and yet ideally stop short of replacing it entirely. I now find myself using it several times a day, along with various other A.I.-driven apps and websites, mostly to fill in gaps in my technical skillset. However, this particular post explores whether A.I. could make me a more productive writer.
“Dear Sir or Madam, will you read my book? It took me years to write, will you take a look?”, Lennon & McCartney
Researchers are known for being verbose. I have been involved in many client meetings where my twenty minute speaking allocation was reduced to five after the researcher before me got carried away with what they were saying. After all, why say what you mean in 100 words when you can use 1,000? By way of example, this article would have worked perfectly well without the inclusion of the paragraph you have just read.
And so when asked to compress my views on the outlook for UK real estate, its risks and opportunities into no more than four hundred words for a magazine article, it was not a challenge I took lightly. As luck would have it, I had recently compiled a presentation on precisely this topic for a seminar. I do not typically make speaker notes on my presentations, and so I would have expected to spend half a day extracting the key message from my slides, before editing them into the finished article.
In this brave new world, however, I opted to use A.I., armed with a research budget of £0. I found the website otter.ai that, in less than five minutes, transcribed a practice run of my presentation that I had recorded on my laptop. Remarkably, it converted twenty five minutes of speech into 3,850 words with the kind of accuracy that was beyond that technology just a few years ago. I noted fewer than thirty minor errors, none of which changed the messaging. There were a few more consequential misunderstandings; for example, calling a building ‘unelectable’ rather than ‘unlettable’, and thinking that the Student Housing sector was ‘a cyclical’ rather than ‘acyclical’. Whether these are attributable to the transcription service, to my garbled delivery or my poor quality microphone, we’ll never know. But I was already suitably impressed.
“No 9000 computer has ever made a mistake or distorted information. We are all, by any practical definition of the words, foolproof and incapable of error”, 2001: A Space Odyssey
The next stage was to sit back and let ChatGPT handle the rest, with the prompt:
“Using formal business language, please convert the text below into a c.400 word article on the outlook for UK real estate”
I decided not to edit the otter.ai transcription beforehand, to see how ChatGPT dealt with the errors and casual phrasing. After all, the test was to be as hands-off as possible.
The results were delivered less than thirty seconds later and were, frankly, outstanding. There was nothing in the article that I disagreed with. And it had even reorganised my original linear approach to story-telling that I find works better for presentations, into a non-linear one that works better for a magazine article. The emphasis of the article was a little off, with too little focus on the economic outlook and too wordy a conclusion. But a couple of simple prompts later, and ChatGPT had amended this to my liking.
“Take a sad song and make it better”, Lennon & McCartney
The most concerning aspect of the exercise was the approximately one quarter of the text generated was not based on the transcription, even though my explicit instructions were to use the text provided.
“The real estate market faces the risk of a credit crunch, limiting credit availability for property transactions,” I had said, although not in so many words. “However, this presents an opportunity for astute investors,” adds ChatGPT. “As property loans worth approximately £40 billion are set to be renewed, motivated sellers may emerge,” I had continued, “providing opportunities to acquire properties at potentially favourable prices,” concluded ChatGPT. It was almost as if it were turning my bleak warning about the risks inherent in real estate today into a sales pitch. Or perhaps it was interjecting balance into an article where it had expected to see it.
Even more interestingly, I had spent seven minutes discussing net zero carbon towards the end of my presentation, and not once did I mention the repurposing of buildings into other use classes. And yet ChatGPT had introduced the idea into this section, and returned to it several times as if it were a major theme. Whilst repurposing is certainly a viable investment strategy – separate to the retrofitting opportunity that I had majored on in the presentation– I had to wonder where ChatGPT had taken it from. Not only did I remove this reference, but it led me to re-read the article it had generated, this time much more closely.
“I did it my way”, Frank Sinatra
Whilst I was impressed with the results, my intention had always been to heavily edit the reimagined version of my own text that I was given. By the time I was finished, I had meaningfully altered about forty percent of ChatGPT’s response. Overall, I had certainly saved myself some time, and an experienced user of A.I. would no doubt have saved even more or reined in the more rogue aspects of the A.I..
And yet it is crucial that the technical tools I use to arrive at or communicate my opinion should not unduly influence it. And so the introduction of unsolicited views into ChatGPT’s responses means that I am unlikely to rely very much on this shortcut for writing in future. At least, not without a very careful eye on the messaging.
That said, a single case study is not a strong enough argument to draw sweeping conclusions. And if the speed at which ChatGPT 4 trounced its predecessor is anything to go by, then many of the concerns I’ve outlined above may have already been resolved before I publish this article. The shelf-life of this post is most likely very short indeed.
“I would have written a shorter letter, but I did not have the time”, Blaise Pascal
Where today’s A.I. could perhaps be better employed, however, is for a company’s Marketing department to take a report or presentation produced by their Research team, and use A.I. to repurpose the messages for a broader campaign: a short form version of a report to be sent to prospects, an even shorter version to be posted on its website in front of the firewall, a video with voice-over-images summarising the key messages to go on LinkedIn and, why not, maybe even a series of tweets. Rather than produce all of these outputs, and get thoroughly bored of repeating themselves whilst doing it, the Research team could instead simply review the various outputs that had been generated for them in no time at all.
“Perfection is achieved not when there is nothing more to add, but when there is nothing left to take away”, Antoine de Saint-Exupéry
To go full circle, I asked ChatGPT to summarise the above commentary in a single short sentence. Here was its response: “Using AI for transcription and content generation saves time but requires careful editing.”
Great observations and well written, Simon!