GPT-3 Generated New York Times Opinion Headlines are Pretty Funny

I used GPT-3 to generate NYT opinion headlines and was equally shocked and amused when I read them

I am a big fan of GPT-3. The content it has generated has been impressive to say the least. It has made me laugh and has left me in awe at the current state of artificial intelligence engines. I have used it sporadically since I was granted access, but have been wanting to use it on a small project to get a real feel for its creativity.

I’m also a big fan of the New York Times. Even though I’m Canadian, I read it religiously for world news and updates on the most captivating drama of all time: American politics.

There was an opinion article that made it’s way on Twitter the other day about Elizabeth Holmes.

While I don’t have an opinion on the article headline either way, tech Twitter seemed to vocally decry the article as just another bad take from the NYT opinions desk.

They are known for having some pretty “out-there” takes. Take, for example, the op-ed from Senator Tom Cotton during mass protests sparked by the killing of George Floyd.

So yesterday when Twitter user Jon Markland (@jonmarkland) posted this NYT opinion article spoof, it went viral, getting over 1,000 retweets and almost 10,000 likes.

The quote is from an Orson Welles film “The Third Man” (I had to look it up). It’s obviously a cynical take on human life, but the message almost seemed apt for the sometimes counterculture remarks of the NYT opinion section.

It was fun to see this tweet take off, so I wondered what GPT-3 could add to the mix.

I created two models: one that simply generated article headlines without examples (only the prompt “This is a list of AI generated fake new york times opinion article headlines:”), and the other based on 125 recent NYT opinion headlines as examples using the same prompt.

The results of the first model were a mix of hilariously random statements and horrifyingly offensive takes. I’ve omitted the latter, but included some of the more amusing ones below in NYT opinion headline format.

And, my personal favorite:

Apart from the obvious ones, they maintained some of the classic edgy contrarian takes that we see from time to time in the NYT.

The other model’s output was actually pretty boring.

After feeding in 125 of the most recent headlines, GPT-3 generated the following:

“The Strange, Sad Decline of the Anti-Vaxxers”

“The U.S. Should Be a Global Leader in the Fight Against Climate Change”

“The Covid Pandemic Is a Global Emergency”

They seem to follow collective public opinion rather than the contrarian nature of the former model.

What does this say? I’m not sure.

Perhaps it says that the NYT opinion section isn’t as contrarian as we think. When GPT-3 used the previous headlines as guidance, it generated some very straightforward agreeable headlines. When I left it to generate the text without prompts, it was definitely edgy to say the least.

In any case, it was a fun experiment. I started posting the less edgy takes on twitter under @not_nytopinion as a memento. I’m going to try to keep generating headlines every day and will post them there.

Hopefully they don’t start reconciling with our reality anytime soon. Though I definitely believe that A.I. could beat Kim Jong-un in chess.