Ben Zhao
speaker
105 appearances
2 recordings
2 series
first heard Jan 2025
last heard Mar 2025
Ben Zhao’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
And that was just mind-blowing because I had been talking to our students for months about having to build for this dark scenario. And literally, here's the New York Times saying, yeah, this is today and we are already in it. That was disturbing on many fronts, but it did make writing the paper a lot easier. We just cited the New York Times article and said, here it is already.
Clearview AI is funded how? It was a private company. I think it's still private now. It's gone through some ups and downs. Since the New York Times article, they've had to change their revenue stream. They no longer take third-party customers. Now they only work with government and law enforcement.
Fox was designed as a research paper, an algorithm, but we did produce a little app. I think it went over a million downloads. We stopped keeping track of it, but we still have a mailing list, and that mailing list is actually how some artists reach out.
These companies will go out and they'll run scrapers, little tools that go online and basically suck up any semblance of imagery, especially high quality imagery from online websites.
It would download those images and run them through an image classifier to generate some set of labels and then take that pair of images and their labels and then feed that into the pipeline to some text image model.
How meaningful is that? Well, opting out assumes a lot of things. It assumes benign acquiescence from the technology makers. Benign acquiescence, meaning they have to actually do what they say they're going to do? Yeah, exactly. Opting out is toothless because you can't prove it in the machine learning business.
Even if someone completely went against their word and said, okay, here's my opt-out list, and then immediately train on all their content, you just lack the technology to prove it. And so what's to stop someone from basically going back on their word when we're talking about billions of dollars at stake? Really, you're hoping and praying someone's being nice to you.
A big part of their misuse is when they assume the identity of others. So this idea of right of publicity and the idea that we own our faces, our voices, our identity, our skills and work product, that is very much a core of how we define ourselves. For artists, it's the fact that they take decades to hone their skill and to become known for a particular style.
So when that's taken against their will without their permission, that is a type of identity theft, if you will.
Right now, many of these models are being used to replace human creatives. If you look at some of the movie studios, the gaming studios, or publishing houses, artists and teams of artists are being laid off.
One or two remaining artists are being told, here, you have a budget, here's mid-journey, I want you to use your artistic vision and skill to basically craft these AI images to replace the work product of the entire team who's now been laid off.
Poison is sort of a technical term in the research community. Basically, it means manipulating training data in such a way to get AI models to do something perhaps unexpected, perhaps more to your goals than the original trainers intended to.
Glaze is all about making it harder to target and mimic individual artists. Nightshade is a little bit more far-reaching. Its goal is primarily to make training on internet scrape data more expensive than it is now.
perhaps more expensive than actually licensing legitimate data, which ultimately is our hope that this would push some of these AI companies to seek out legitimate licensing deals with artists so that they can properly be compensated.
We're talking about companies and stakeholders who have trillions in market cap the richest companies on the planet by definition. So that completely changes the game.
It means that when they want things to go a certain way, whether it's lobbyists on Capitol Hill, whether it's media control and inundating journalists and running ginormous national expos and trade shows of whatever they want, nothing is off limits. That completely changes the power dynamics of what you're talking about. The closest analogy I can draw on is in the early 2000s, we had music piracy.
Folks who are old enough remember that was a free-for-all. People could just share whatever they wanted. And of course, there were questions of legality and copyright violations and so on. But there, it was very, very different from what it is today. Those who are with the power and the money and the control are the copyright holders. So the outcome was very clear.
exactly you had armies of lawyers when you consider that sort of situation and how it is now it's the complete polar opposite meaning it's the bad guys who have all the lawyers
Well, I wouldn't say necessarily bad guys, but certainly the folks who in many cases are pushing profit motives that perhaps bring harm to less represented minorities who don't have the agency, who don't have the money to hire their own lawyers and who can't defend themselves.
These companies are basically exploiting the fact that we know lawsuits and enforcement of new laws are going to take years. And so the idea is, let's take advantage of this time. And before these things catch up, we're already going to be established. We already are going to be essential and we already are going to be making billions.
Showing 21–40 of 105 · page 2 of 6
← Previous
Next →