My early take on ChatGPT

Late in August, our eldest convinced me to try ChatGPT. I was, frankly, skeptical. Three weeks in, I’m a convert. I’ve achieved more in that time than I would normally in 4-6 months (actually). Here’s my earlier take on the good and not-so-good elements of ChatGPT.

First, let me give you sense of how I’ve used it:

• Fantasy world to rival all others: Those of you?who know me appreciate I like Young Adult fantasy - think Harry Potter, Dark Materials, Eragon. I’ve always promised myself I’d write a series, but work and life has stood in the way. Well, now I’ve created a world I’m very excited about. I’ve outlined a 10-book saga, 5-6 prequels (including a graphic novel), 4-5 novellas and three sequels. I even know the smells, colors, food, drink and clothing for the time the adventures are set, as well as the politics, culture and key figures.? Oh, I also have ideas for picture books, early reader books, cook book, board game, video game, musical and much more. Yes, it’s all very high level. But with a solid two week’s work, I’m confident I can finish chapter outlines for all 10 books. In practice, I’ve already done 1-2 years worth of complex work.
• Series of guides for directors: I challenged myself to author 30 articles on aspects of governance in 30 days. I’ve already published 12 - you can find them on my LinkedIn profile. Three more are almost complete with more to come. For sure, I’m generally a fast writer (on a good, coffee-charged day I can write 10,000 words), but I’ve never authored 15 articles in 10-11 days.
• Complex business opportunity: Without giving anything away, I’ve been providing some input to a friend on an exciting business opportunity in a sector and place that’s entirely new to me. It’s required a lot of research on business models, financial projections, potential partners, legal matters and much more. It would have taken me six months to do what I’ve just done.

In short, a lot for three weeks.

So what are my early impressions? First, the positives.

• It’s awesome for desk research: As the above shows, ChatGPT makes light work of research. Even simple questions evoke good initial structured responses. With better questions, come better results.
• Its early drafts are quite good: With a few iterations and direction, ChatGPT can draft good content. In a world where we depend heavily on frameworks, policies, standards, guidelines, proposals and so on, ChatGPT can get you to a good draft - say 75% complete - in a matter of hours.
• It’s at its best when you engage it in dialogue and show it respect: ChatGPT doesn’t know everything - in fact, one can argue, it doesn’t know anything;? it’s just good at learning from others and making their thoughts more cogent. But who cares how it ‘knows’ what it knows. Just go with it. Talk to it like you would another smart cookie.
• It never tires: My brain is alert when it’s alert. Alas, it means, oftentimes, when I want to engage a colleague in detailed discussion, they aren’t available. Without ChatGPT, one has to take notes to remind oneself what you want to talk about later. With ChatGPT, you simply open the app and off you go. It doesn’t need a coffee to wake up and it’s not distracted by other matters (though, per below, it can be overrun); it’s just ready to spar 24/7.
• You can regenerate answers: If you don't like a response, but feel your question should be good enough to get a useful response, just click regenerate. It's not offended. But beware, if you go back up your chat string and edit a question some ways back to get a new response to a prior question, it'll delete all subsequent questions and answers; so copy those before you go backwards and submit a revised question.
• It’s delightfully polite: I don’t know about you, but in my experience when a colleague has made a mistake, they often find it hard to apologize or do so without conviction. ChatGPT was, by contrast, taught manners by my grandmother. It routinely apologies when it misunderstands your requests. It also often says “That’s a great idea!” What’s not to love?

So, the shortcomings:

• It doesn’t know everything: This is probably my most profound learning. ChatGPT has incredible breadth. I'm sure there's subjects it struggles with, but I haven’t found them. But its generative nature means it can only get you so far. Without some knowledge of the subject matter, it will be hard to push it to really high-quality results. One needs to know when the content is wrong or written in a confusing manner. That 75% draft required work from you, to push it on content, structure and style - but the other 25% is all down to you. It’s futile to think it’ll get you to 99%, at least, not yet.
• It’s forgetful (in effect, by design): ChatGPT was trained up to September 2021 and launched November 2022, so let’s say, in AI years, it’s 15 years old. Alas, its generative nature means it comes over as forgetful as a 110-year old. I’ve done some really long chat strings with it and it can forget detail of what we said four days ago in the heat of a detailed debate. That’s OK; it’s almost human. But it forgets what it said three responses ago and that’s not OK. You have to keep reminding it what it said - it grates after a while. (You should see how the three characters evolved in my 10 books - the characters’ eye colors even changed!)
• It’s writing style can be beyond cliched: Since it’s drawing from the ether, it can write in a way that makes a business consultant seem like Shakespeare. It uses cliches, analogies and juxtapositions that would distract any reader. You have to describe the style you want or, even better, give it an example to mimic. It doesn’t get it perfect, but without direction, don't expect good prose.
• Its grammar is horrible: For someone who disliked English at school, I’m fairly militant on grammar. It matters. ChatGPT has some very annoying habits it can’t kick. To name a few: with one exception (below), it puts commas everywhere (yes, I dislike serial commas); it uses the construct “not only does one thing need doing, but so does another” way too often; it loves hyperbole, such as "holistic," "comprehensive" and "unique"; it is addicted to the verb "ensure," even when you tell it not to use it; it unnecessarily drops “the”, “that” and “also” in like sprinkles; it doesn't know a comma should precede a "but"; it can’t hyphenate well at all.
• Exporting data is painful: The cut-and-paste feature is primitive. Its ability to create tables is also painfully basic. Get ready to do a lot of reformatting; 'find-and-replace' in whatever software you use will be your new best friend.
• It’s in high demand: While it never sleeps, it can be pulled in too many directions. Be prepared for the odd time it tells you it’s not available due to over-demand. But don’t worry - those amazing engineers at OpenAI are on it and it’ll be back soon.

My overall take: ChatGPT, where have you been all my life? I can see why some people are worried AI-based applications may greatly impact their professions and lives (think the current writers’ strike). But I’m a glass-half-full type of person. With ChatGPT, I’m confident I’ll achieve more in the next 10 years than I did in the first 30 years of my career. That’s exciting, energizing and, yes, a little frightening.

The views in this article are mine. Copyright: Mark Watson