😺 AI Katy Perry tricks the world

Welcome, humans.

AI companies are so into sponsoring The Neuron that all our Q2 primary ad spots got snapped up weeks ago. Sheesh!

Lucky for you, there's still a chance to showcase your product to over 430,000 AI power users via secondary slots in ‘Treats To Try’. Jump on this opportunity before it's too late!

Here’s what you need to know about AI today:

• An AI deepfake of Katy Perry fooled millions.
• OpenAI is building an AI image detection tool that’s 98% accurate (so far).
• OpenAI might postpone its big product event to next Monday.
• Wayve, a startup building self-driving cars, raised $1B+.

On Tuesday’s podcast: OpenAI’s New Competitor?, MAI-1 and Synthetic Data, Mustafa Suleyman (Apple Podcasts, Spotify, YouTube).

AI deepfake strikes again.

Ready for some AI drama that’s absolutely fireworks?!

Yesterday, Katy Perry posted AI-generated images on Insta pretending she was at the Met Gala. This fooled A LOT of people, including her own Mom, who texted, “Didn’t know you went to the Met”.

We don’t know what’s more ridiculous: the fact Katy’s Mom was duped by a photo of her own daughter OR millions believing that Katy’s dress, which was so absurd you’d think it was designed by the fauns of Narnia, was legit:

Even more ridiculous is the fact that this isn’t the first time millions have been duped by an AI photo—remember that photo of The Pope sporting a Balenciaga puffer that was created with Midjourney?

Well, that happened over 1 year ago (!), and social media platforms still haven’t figured out how to tackle deepfakes.

Sure, they’ve promised to tag all content that uses their AI tools, require creators to disclose when content uses AI, and even started labeling images from other companies’ AI.

Yet, it falls short time and time again…

A Reddit screenshot reveals one AI Katy Perry image hit at least 9M views on Twitter before a community note clarified, “This is not a genuine photo of Katy Perry at the 2024 Met Gala.” #TooLittleTooLate.

Another potential option: OpenAI just announced a new tool that can detect images created by its AI, DALL·E 3.

The tool correctly spots 98% of DALL·E 3 images and 99.5% of real images.

Impressive, but not perfect. The tool's accuracy dips when DALL·E 3 images are modified with things like Gaussian noise.

And with 98% accuracy, that means 2% of AI photos might slip through, potentially letting billions of fake photos flood our feeds undetected.

• It’s like those AI detection tools that claim to detect 99% of AI writing.
• With 1% false positives, millions could be wrongfully accused of using AI (& many have been). It’s just not reliable enough…

For now, the rule is, “If you see something that looks off, say something.”

FROM OUR PARTNERS

Want to invest in a Banksy? Get 5 free shares ($100 value) with this exclusive offer.

For a limited time, the art investing platform Masterworks is offering new users 5 free shares, in the form of $100, when they sign up – no strings attached. All you have to do is create an account, and complete your onboarding call. It's that simple.

With this offer, there may never be a better time to join more than 64,631 investors discovering the power of investing in shares of blue-chip contemporary art by artists like Banksy, Basquiat, and Picasso.

Benefits include:

• High historical appreciation, with Postwar and Contemporary art prices outpacing the S&P 500 by 64% from 1995 to 2023
• Portfolio diversification potential, contemporary art prices have shown very-low, negative correlation to the S&P
• Proven results, with 3 representative sales, Masterworks investors realized net annualized returns of 13.9%, 14.6%, and 17.8%.

This is a limited time offer, so if you’ve heard of Masterworks but never took the time to learn more, now's your chance. Just use this special link to get started.

Around the Horn.

• New developments from OpenAI (it’s their world, we’re just living in it):
  • OpenAI might postpone its big product event to Monday, where there’s speculation it could announce an AI search engine.
  • OpenAI is partnering with big publisher Dotdash Meredith to integrate content from its 40+ brands into ChatGPT.
  • OpenAI COO Brad Lightcap said that today’s AI will look “laughably bad” within a year, and future models will shift toward verbal interfaces (think: speaking to ChatGPT).
  • OpenAI and Microsoft launched a $2M fund to fight AI election deepfakes.
  • OpenAI is developing a tool called Media Manager that will enable content creators to opt out of having their content used to train AI models.

Wednesday Wirings.

1. Wayve, a builder of self-driving cars, raised $1.05B.
2. DocuSign acquired AI-powered agreement management software Lexio for $165M.
3. Lamini, which helps enterprises build AI models on their proprietary data, raised $25M.
4. *Get your company featured here in front of 435,000 AI power users. Fill out this form ASAP before slots fill up!
5. Daloopa, an AI platform that extracts and organizes data from financial reports, secured $18M.
6. Espresso AI, which helps companies reduce Snowflake cloud costs with AI, raised $11M.
7. Sagetap, which uses AI to match buyers and sellers of enterprise software, snagged $6.8M.

*This is sponsored content. Advertise in The Neuron here.

A Cat's Commentary.

Where do you #Neuron?

Send us a picture of where you read The Neuron here (with some spicy commentary), and you might get featured next week!

Shoutout to this week’s #Neuron Champ: Legend says this cat was automated by AI eight lives ago and now enjoys its days resting and reading. A life well lived!

Sai and Ty write (from Brooklyn, New York): First newsletters, then birds.

Eloy writes (from Meran, Italy): Feeling like a Neuron in the middle of mountains.

Joshua writes (from Nassau, Bahamas): Love my office!

Ilia writes (from Tbilisi Georgia): Ilia from Tbilisi, Georgia. :)