The Changing Conversation Around Gen AI

I was 10 when my parents took me to Foot Locker to get my first pair of Air Jordans . I had convinced them that the $65 was worth it. And I had convinced myself that they would make me a better basketball player.

Turns out, it wasn’t the shoes .

They didn’t make me jump higher, run faster or shoot better. I was just another suburban kid with a fancy pair of footwear left to reckon with his limitations on the court.

Today, I can’t help but think that many companies saw Gen AI as Air Jordans. No doubt, they believed that, with some clever prompt engineering, they could use it to crank out killer content that would drive sky-high traffic.

By now, that dream is fading. They now see a sea of Air Jordans and realize they’re not special.

The realization is healthy. And it seems to be changing the conversations around Gen AI and content creation. It's less talked about as an all-powerful plug-and-play solution. It’s much more talked about as a transformational tool that has limits.

That’s not to say its prevalence is diminishing, of course. In fact, I’ve noticed an increasing number of people willing to share their Gen AI usage. It’s not quite the same taboo it was just six months ago.

In the writing process, use cases are apparent. The summarization of research, for example. In a matter of seconds, we can get bulleted takeaways from complex research papers. I have also enjoyed getting feedback from AI on early drafts.

But as we continue to explore what Ethan Mollick calls AI’s “jagged frontier,” we see its weaknesses and biases.??

Something missing

The novelist Curtis Sittenfeld’s writing contest in the New York Times against ChatGPT illuminates those limits. With the same prompts, Sittenfeld and ChatGPT each wrote a 1,000-word short story based on five prompts: lust, kissing, flip-flops, regret and middle age.

Take the time to read both. It’s obviously which one Sittenfeld wrote. In a podcast interview , she pointed out one big difference: her writing had soul. ?

“I feel like if you plucked an individual sentence from ChatGPT’s story, it’s probably fine, or even better,” she said. “It’s probably smoothly written, but cumulatively, something is missing. There’s not any real emotion or anything that makes it feel very specific.”

Most of us are not writing best-selling novels. But for those of us who toil in creating copy for businesses, we can learn from Sittenfeld: There is no shortcut to making that critical human connection with readers.

The flood of AI-generated content on social media reminds me of the over-reliance on SEO, which is responsible for filling data centers with passable but unmemorable content. Rather than starting with good ideas and valuable insights, digital marketers believed the key to editorial success was the right keywords.

But it turns out, algorithms prioritize content quality. Trying to game the system without a commitment to editorial excellence was a fool’s errand. It still is.

AI slop

The flood of Gen AI has a larger problem for those companies depending on it to stand out: AI slop. That’s a term for AI-generated text that has been trained on AI-generated text.

As The New York Times reported this week , it’s already a concern. Research has shown that AI models that continue to ingest AI-generated text produce increasingly worse and less diverse results.

Gen AI relies on large data sets used to predict the sequence of a sentence. While it’s very good—and will only get better—the results can often have a bland, safe, sameness quality. But if AI texts infiltrate the underlying data sets AI trains on, the results can become even more similar, less useful and more susceptible to bias.

To edit is divine destiny?

Gen AI is not going away, of course. It will continue to transform how we work. But, for now, it requires constant editing and oversight for most of its output. And that can be a powerful combination.

While ChatGPT did not write a better beach read than Curtis Sittenfeld, it did write its story a gazillion times faster. Could a skilled fiction writer have edited it so that it could compete with Sittenfeld’s soulful rendering? Maybe.

I keep thinking about a prediction made by Ezra Klein that AI is going to turn many of us into editors. ?The efficiencies of giving drafting responsibility to AI will be too hard to ignore. But, as he suggested, there will be a price.

“You make more creative breakthroughs as a writer than an editor,” noted Klein. “The space for creative breakthrough is much more narrow once you get to editing.”

This week, Nvidia, the chip maker fueling the AI economy, reported nearly $20 billion in operating revenue for the most recent quarter, more than double the same quarter last year. The market rewarded it by sending the price of shares down.

Perhaps the hype around AI is too much to live up to. Maybe that’s not a bad thing. Maybe now we can focus on using it without losing ourselves and our creative edge in the process. ?