Welcome to The Neuron.
Today is January 26th. If you're still saying "Happy New Year!", we officially give you permission to stop.
Today in AI:
- OpenAI's AI Detection Plan Put to the Test
- Oops: Robot Lawyers Can't Practice Law
- Shutterstock Now Generates AI Images
- An AI Tool For Your Weekly Updates
OpenAI's AI Detection Plan Put to the Test
Teachers everywhere breathe a sigh of relief.
OpenAI has hinted that they want to "watermark" ChatGPT output to curb online spam and give teachers a way to catch students submitting ChatGPT work as their own.
The watermark would be hidden from humans but a special computer program could somehow pick up on it. Crazy math, yadda yadda, you know the drill.
Researchers at the University of Maryland decided to put the idea to the test and confirmed that it'd probably work.
Why can't you just swap out words? You'd have to swap out 40-70% of the watermarked words while still making it sound good to escape detection. Possible, but most people will get caught.
Pretty much any tool would be able to build in the detection formula. Which means students using ChatGPT to pass World History don't have much time left to run circles around their teachers.
AI giveth, AI taketh away.
Oops: Robot Lawyers Can't Practice Law
You're safe for now, lawyers
DoNotPay tried to send its AI lawyer to a February traffic court case. It got a lot of attention - most of us were like, "Cool! I wonder if it'll win!"
It also summoned the fury of a 1000 state bar associations, which came raining down on DoNotPay and its CEO Joshua Browder after the media picked up on his shenanigans.
After enough letters threatening criminal charges, prosecution and potentially prison time, Browder backed off.
Lesson learned, Joshua: Never get in the way of a lawyer and their ability to bill hours.
DoNotPay is now focusing on things like fighting medical bills and subscriptions. But if it didn't win an actual court case, it still put lawyers on notice.
An AI of some kind is coming for the legal profession, whether they're ready or not.
Shutterstock Now Generates AI Images
While Getty Images sues, Shutterstock leans in.
Shutterstock is owned by Adobe, which has made it clear that they're leaning into the generative AI experiments:
- Adobe Express (their Canva competitor) lets you generate content using AI
- Adobe Stock (one of their stock photo sites) lets you upload AI-generated images
- Now, Shutterstock has an AI image generator built on OpenAI's DALL-E
Part of the leaning in is recognizing the artists that make the AI possible. Artists are not very pro-AI right now - some are suing for copyright.
Fork over the cash. Shutterstock has set aside the Shutterstock Contributor Fund to pay anyone who's images were a part of the AI's training set. Anytime they make money from generative AI, you get some money.
Around the Horn
- Do you write end-of-week updates at work? Try Broadcast - it drafts one for you in one line using AI. People on Twitter seem to be big fans.
- Andi: AI-powered, chat-like search
- WhisperGPT: Automated phone calls using AI
- Using AI to censor a film. This was a pretty mindblowing clip.
- Atomic AI raises $35 million to discover drugs using AI
- Five pieces of advice for those building in AI right now
- CNET got in big trouble for publishing AI-written articles that turned out to have factual errors. Here's their "apology" (read: we're going to keep doing it)
DM me links on Twitter: @nonmayorpete
Are you new to all this AI stuff? Here's The 3-Minute Guide to Slaying Your Dinner Convo About AI to get you up to speed. Or at least smart enough to impress your family.
