The big reveal
But now the important part: comparing the AI-generated headline(s) with the hand-crafted artisinal human-generated ones. How do the suggestions compare with the final headline the NYT used? 1 Let’s take a look:
The actual New York Times headline is far more vague, less fear-inducing, and doesn’t mention the Bronx at all. It makes it a story relevant to all New Yorkers, but doesn’t turn it into fearmongering clickbait.
Honestly, I was shocked: it’s so far from anything that GPT suggested. Maybe your prompts got you closer, but I was certainly not that lucky.
Reflection
This whole process made me think about a paper comparing story generation between humans and humans assisted by AI: 2
Stories written with AI assitance were scored as better than stories written by humans alone. Despite being worse, though, the human-writtens stories had a much wider range of ideas and concepts in them. AI guidance provided the authors a higher-scoring work, but at the expense of actually being unique!
To abuse a quote: “all human work improved by AI is alike; every purely human work is creative in its own way.”
In future chapters we’ll dig deeper into how AI tools like ChatGPT work, and how a side effect of their construction might mean that headline generation is the only thing they’re reasonably good at! So whether you were super excited or roundly unimpressed: get excited for a ride down to the bottom.
Footnotes
Again, big shout-out to Sisi Wei of The Markup for this one, which she suggested during a panel at ONA23. Seeing how much better the human-generated headlines can often do a fair job shocking you out of your ❤️ of AI-geneated headlines.↩︎
I can’t find it! Only this one that is only kind of related. I promise it’s real, though: if you let AI help you, you’re probably more boring than an actual human (although yes, your story will be better written).↩︎