Rao said Adobe XD, Adobe's earlier product design tool, made only about $15 million in standalone revenue before being put in maintenance mode.
“It was only making $15 million of standalone revenue.”
Person
Dana Rao
1 episodes · 23 quoted claims
Adobe launched its Firefly foundation model for text-to-image generation in March of the year before this interview.
“last March, we launched our foundation model Adobe Firefly, you know, which is helps uh people do text to image generation.”
Rao said his organization at Adobe has about 700 people, split roughly between about 350 in security and about 325 in legal and public policy.
“My whole org is 700, so it's about fifty fifty between legal slash public policy and then security, so security is about 350, and legal is probably about 325”
Rao said Adobe built a security sandbox around Acrobat so that even a discovered vulnerability would be hard to exploit.
“acrobat is a great example of a place where we dedicated security resources to making sure there's a sandbox essentially around Acrobat.”
Rao said Adobe runs an anti-piracy effort because pirated copies of its software are widespread and often outdated and insecure.
“Because there's a lot, as you know, a lot of pirated Adobe software out there.”
Adobe was founded by Chuck Geschke and John Warnock, and Rao said Warnock had died a couple of months before the interview.
“we had two amazing founders, um, Chuck Geschke and John Warnock. John just passed away a couple months ago”
Rao said the Adobe-Figma deal entered the in-depth phase two review with both the UK and the European Commission in June and July.
“as soon as you go into phase two, which we went into in July, June and July, with both the UK and the European Commission.”
Rao said the UK and European Commission decisions on the Adobe-Figma deal were due on February 5 and February 25.
“And in our case, the decisions were February fifth, February twenty fifth.”
Rao said Figma had about 800 employees and argued that regulators wrongly assumed it could expand into many markets at once.
“Figma's gonna build Photoshop or whatever, they have 800, 800 people, but then they're gonna somehow magically do everything right.”
Adobe trained its Firefly image model on Adobe Stock content and other licensed work instead of scraping the web.
“we decided to train Firefly on our own Adobe Stock content and other licensed work from the rights holders to build our model.”
Adobe launched the second version of its Firefly model at its Adobe Max conference, about a month before the interview.
“we were even more excited about the 2nd version of the model, which we launched at Adobe Max.”
Rao said the EU AI Act, passed in a first stage, says AI developers must obey the EU copyright directive, which may limit training on web data in Europe.
“But even in there, they said that you have to obey the EU copyright directive.”
Rao argued that generative image models misspell text because they treat letters as symbols, which he cited as evidence they don't copy the images they were trained on.
“it always gets text wrong, generative AI always gets text wrong, cause it has no idea it's words.”
Rao said the Supreme Court's Warhol decision showed the Court cares about economic harm in judging fair use.
“The Warhol case clearly showed the Supreme Court is interested in the economic factor as part of the fair use analysis”
Adobe wrote a draft federal anti-impersonation bill and gave it to the Senate Judiciary Committee for review.
“we wrote a draft bill, we gave it to the Senate Judiciary Committee. They're reviewing it.”
Rao cited a deepfake of Ukrainian President Volodymyr Zelenskyy telling troops to lay down their arms as an example of harmful AI misuse.
“But if someone is actually trying to create harm, like there was a, you know, image of of Zelenskyy saying, you know, lay down your arms, right?”
Rao said Adobe's AI scientists found that deepfake detection reached only about 60% accuracy, which was not good enough for consumers.
“And the answer came back, it's like, probably like 60% accuracy, so not consumer grade.”
Rao said Leica was already selling a camera that can attach Content Credentials metadata at the moment a picture is taken, and Sony had also announced such cameras.
“Leica is already shipping that camera, it's in the stores. And that means you can take a picture, turn on Content Credentials”
Rao said attendance at the Content Authenticity Initiative symposium grew from 59 people at the first one, four years earlier, to 180 at the latest at Stanford.
“the first time we did this four years ago, there was fifty nine people. This time, there was 180 people.”
Rao said the C2PA standards group produced version 1.0 of its provenance specification after a year and had reached version 1.4 after four years.
“After just a year, we had the 1.0 version of that technical specification, and we're already up to 1.4 in just four years.”
Qualcomm announced that its Snapdragon chips would support the C2PA content provenance standard.
“Qualcomm just announced that Snapdragon is going to be C2PA compliant.”
Adobe's Content Authenticity effort offers a verify site where users can upload an image with Content Credentials and use a slider to compare it with the original.
“we have a site called verify dot org, and on that site, if you bring your content credentialed image, you can just use a slider.”
Rao said every Adobe engineer who creates an AI feature must fill out an AI impact assessment, with higher-risk features sent to an AI ethics review board.
“So everybody who creates an AI feature at Adobe has to fill out an AI impact assessment, every engineer”