Hi, welcome to your Weekend.

For about 11 months after OpenAI released its text-to-image generator Dall-E 2 in April 2022, many of us wondered when Adobe would come to market with a rival product, which could be neatly integrated into design programs like Photoshop, Illustrator and Premiere. 

The wait felt long, especially considering how rapidly startups like Open AI, Stability AI, Midjourney and Runway ML were shipping mind-boggling new tools. Finally, this March, we got what we’d been waiting for when Adobe launched a beta version of its text-to-image model Firefly. The program created instant buzz in design and AI circles—and convinced Margaux to delve into this week’s cover story about Adobe, the no-longer-sleeping AI giant.  

For evidence of how cool and easy to use Firefly is, just look at the images adorning this newsletter, all of which were created by designer Clark Miller with the assistance of Firefly’s generative fill tools in Photoshop. Here’s Clark’s unofficial beta review of Firefly:

“Biggest strength for me is that it’s built into Photoshop so it fits really well in my workflow, unlike Dall-E and Midjourney, which force you to work within their respective web programs. It’s also not limited by any resolution restrictions, whereas other generative AI programs output a relatively low- or medium-resolution final image. Firefly matches the quality and resolution of the image you’re working on.

“Another big plus: since you’re starting with a base image you want to build off of (which could even be an AI-generated image), you don’t have to tell Firefly everything you want in the final image all at once. There are entire industries being built around Midjourney prompt engineering, that teach you how to properly construct a prompt so it includes everything from the style to composition to lighting to detail level, etc. With Firefly, I can slowly build an image with a prompt here and a prompt there and then maybe a stock element, or my own illustration, then ask it to blend all of the above—or build off of what I made just in certain areas of the image.

“The biggest cool factor is that I still feel like I’m creating something instead of just asking AI to make it. It’s much more of a collaborator in that regard, helping me along the way instead of taking control of the entire output. Maybe that’s just self-deception on my part, but since it’s just another tool in Photoshop, I feel more like I’m using it to help my work rather than giving the AI full authorship.”

Now onto this week’s stories...

the big read

Can Adobe Catch the AI Bug?

With its $20 billion Figma acquisition imperiled, the software giant, guided by product czar Scott Belsky, is turning its attention to Firefly—its copyright-safe generative AI model. Margaux asks Belsky whether “generative fill” can help the company get its groove back. 

the new ballgame

The Messi Model: How One ‘Extraordinary’ Contract Could Shift the Power Balance in Sports

Lionel Messi, the Argentine soccer maestro, is set to sign a contract to join MLS’s Inter Miami CF that is remarkable in its scale and ambition. Daniel Kaplan explores how this deal—which takes inspiration from how tech companies compensate star employees—could give players unprecedented leverage.

market research

Tech on the Trail: Silicon Valley’s Favorite Smart Hiking Gear

Today’s hiking accoutrements aren’t your basic lace-up boots and walking sticks. New high-tech upgrades can improve hikers’ comfort and safety, plus make the whole experience a bit more fun. Beth Shapouri spoke with some of Silicon Valley’s most avid hikers to see what they’re using to elevate their time on the trails.

Generating: China’s AI chip black market 
There’s a new product flooding the shelves of China’s black markets: Nvidia AI chips. Reuters reports on the aftermath of President Joe Biden’s call for Nvidia to stop exporting its most powerful chips to China. Despite the sanctions, the chips are still easy to find: Reuters was able to talk to at least 10 electronics vendors in Hong Kong and mainland China who said procuring the chips wasn’t a problem. The vendors were able to snatch them from excess stock delivered to US firms or other foreign companies. And the sanctions have given underground vendors an opening to jack up prices to $20,000—double what they normally go for. Even though getting your hands on a handful of Nvidia chips is relatively easy (if pricey), the US government isn’t concerned: to run a powerful AI model like OpenAI’s GPT, a Chinese company would have to amass 30,000 chips—far more, experts believe, than the black market can provide. —Margaux

Watching: “Rich dudes pissing in their pants”
That quote, delivered with manic glee by Pete Davidson, would make a great name for a movie about the GameStop short squeeze of 2020. Instead the producers went with “Dumb Money”—which also captures the inanity of the period pretty well. The newly released trailer for the film is a three-minute tornado that will trigger joyous memories for some, and traumatic flashbacks for others. The film’s credits alone should make you excited for the September 22 release. It’s based on a book by author Ben Mezrich, whose previous work inspired “The Social Network.” It’s directed by Craig Gillespie, best known for the awesome “I, Tonya,” and the cruelly underrated “Cruella.” Screenwriters include Wall Street Journal alums Rebecca Angelo and Lauren Schuker Blum and—get this—Cameron and Tyler Winklevoss. And the cast is studded with stars, including Seth Rogen, Paul Dano, Davidson, Nick Offerman, Shailene Woodley and America Ferrera. This is one Hollywood production you probably should not short. —Jon

Reading: The ever-expanding AI machine
“How many humans does it take to make tech seem human?” reads this week’s New York Magazine cover line. “Millions.” As part of his investigation into the dark underbelly of large-language models, journalist Josh Dzieza spoke with dozens of domestic and international “annotators” who are training, teaching and powering AI. He even took a stab at the work himself. The tasks are, in most cases, mindless and menial, consisting of labeling and sorting thousands of images of shoes and cars and, sometimes, C-3PO. These processes are frequently outsourced to other countries and pay little. “The act of simplifying reality for a machine results in a great deal of complexity for the human,” Dzieza writes. But the story provides a fascinating counter-narrative to many of the fears about AI—that it’ll eliminate jobs—in favor of the notion that perhaps AI creates a different sort of job. One that is “more alien, more isolating and more tedious.” —Annie

Makes You Think

Someone tag @VCBrags. 

Until next Weekend, thanks for reading.

—Jon

Weekend Editor, The Information