15 November 2022
I asked a robot to write a blog post. Here's what happened
I am very bullish on artificial intelligence. I believe that technology as a whole can have a radical impact on humans lives, and it's obvious to me (and everyone else) that intelligent technology is a huge unlock towards that.
I've relied on AI to assist me with content every day for the past two months — every single thumbnail on the Journey blog was generated by DALL-E. Like anyone that has written lengthy content knows, it's really hard to find complementary imagery to content! You end up writing a mass of words and then have to spend a ton of time finding imagery to complement it — and that is before you start struggling with things like thumbnails for social media etc etc.
Whereas I have been able to focus on writing content and then tasking DALL-E with creating complementary imagery, it's been fun, productive and impactful:
Since DALL-E was released, there's been a lot of focus on "AI generating content" — not just images, but text. Of course, OpenAI launched GPT-3 before DALL-E, but it feels like GPT-3 has gotten more usage by startups building on top of it in the past few months.
I also believe that content should be good. Why bother wasting not only the creators time but also the readers time by writing some drivel that is what I call clipart prose. I am proud that all of the blog posts I have written were written by me and are all at least content that I consider interesting!
So I was curious and skeptical about AI driven content. I understand how DALL-E can be used as a almost a graphic designer replacement — instead of finding an upwork designer to create a blog thumbnail, I simply task DALl-E and away we go.
But can AI be a replacement for copywriting? I was immensely skeptical. Writing content is really hard. I can totally see AI being useful as a complement to writing copy — helping with grammar, handy links, research, etc — but I couldn't believe that it could generate content in the way it generates images.
But I wanted to give it a shot. So I did.
I picked this topic because it seemed appropriately vague. I wouldn't expect AI to generate content about nuanced topics like Product-Led Growth (or weather apps!) but a vague and waffle-y topic with a few meaty phrases ("sales people", "sales", "presentations") felt like it would give AI a fair shot.
The product I used generated a few sub topics for the post:
At first glance they seem great, but when you read and re-read them ... they really do not say much. They're very much clipart prose.
Not to be swayed, I asked the AI to generate the entire post:I have no idea why its suggesting "Ethereum, Bitcoin" as topics — presumably AI does not get deployed to suggest keywords, which is ironic.
The content is honestly not great. I do not think this is surprising but I do think its worth explaining.
Whereas DALL-E has to simply generate an image ("a black and white sketch of a mountain") writing dozens or hundreds of words is much harder — it's really hard to have any idea what you want a blog post to say when you're only given a few sentences and then, because you're a robot, be unable to ask questions or go away and research the topic to create a proper feedback loop.
You might argue that I should have given it more guidance — but I think the fact it generated 5 bullet points is more than enough copy for a brief. In fact, I'm going to send this brief to a copywriter to compare. We'll follow up!