Tip: AI You Can Actually Trust

I’ve been warning you not to trust facts from tools like ChatGPT, but Erin Servais just showed me something that changes my mind. It’s called Perplexity AI, and after seeing it, I can imagine never going to Google again.

For all the searches I’ve done, I’ve gotten a high-quality answer — with clear, credible sources.

Notice in the screenshot below that the answer has numbered references that link to source boxes at the top. In the searches I’ve done so far, the top sources are the same sources I get from a Google search on the same question. When I’ve clicked them, I haven’t found any inaccuracies.

Much like Google, it gives you suggested follow-up searches at the bottom or you can enter your own. If you click a plus sign or ask your own follow-up, it generates the answer on the page below the original answer, essentially extending the article it’s creating for you.

A final nice feature of Perplexity is that you don’t have to create an account or sign in. I’m no privacy expert, but I read that if you use this with a VPN (a virtual private network that hides your location and IP address), you have much better privacy than with other systems.

Thanks again to Erin for the tip! Check out her AI for Editors courses.

News

AI ‘cheating’ is challenging and demoralizing professors

theatlantic.com

This sentence crushed my soul: "The teacher wondered why he was wasting his time grading automated work the students may not have even read."

After the first term when students are widely using AI, systems like TurnItIn are turning out to be terrible at flagging cheaters, and students are using systems like ChatGPT in both lazy ways and creative and interesting ways.

And although it's hard to muster sympathy for established tutoring (cough, cheating) companies, they're suffering too. Chegg has lost about $1 billion in market value since saying last month it's losing business to tools like ChatGPT.

‘ChatGPT detector’ catches AI-generated papers with unprecedented accuracy

nature.com

In what would seem to be a direct contradiction to the previous story, researchers say a new system identified 100% of AI-written introductions to academic journal articles. One difference may be the more limited writing style the system needs to process compared to the highly variable type of writing a system for professors needs to process.

The detector uses stylometrics: features of writing style, including “variation in sentence lengths, and the frequency of certain words and punctuation marks” — the same method used to determine that JK Rowling was the author of "The Cuckoo's Calling" back in 2013.

One thing that worries me is that I actually don't care whether a researcher uses ChatGPT to help write the introduction to a paper. I care whether it's accurate. And I worry this will be used to penalize people who speak English as a second language and turn to AI for writing help.

Adobe is selling fake AI images of war in Israel-Palestine

crikey.com.au

Another one for the "Come on, companies, this isn't that hard" file: Adobe is selling realistic AI images of the Israel-Hamas war without any indication they are fake.

This was originally reported by the Australian outlet "Crikey," but it has two triggers that made me skeptical:

1. It’s reinforcing an outrageous thing I already believe: companies are being stupidly irresponsible with AI.

2. It’s about a topic I know is a huge target for misinformation: war.

Therefore, I made sure to find a second credible source before passing it along.* "Motherboard" appears to have done some additional reporting on the topic, and I believe it to be true.

What is an AI sidequest?

Using AI isn’t my main job, and it probably isn’t yours either. I’m Mignon Fogarty, and Grammar Girl is my main gig, but I haven’t seen a technology this transformative since the development of the internet, and I want to learn about it. I bet you do too.

So here we are! Sidequesting together.

If you like the newsletter, please share it with a friend.

^{* #ShowingMyWork}