Jacob Kastrenakes

There’s a lot to discuss — for one, the ongoing chaos of working with Elon Musk and transforming Twitter into X. But also, two events just today.

Twitter’s former trust and safety leader, Yoel Roth, was on stage less than an hour ago discussing how Musk sent harassment his way. Also today, Media Matters came out with a report about X placing NFL ads on accounts of white nationalists. That’s a big ding in Yaccarino’s promise that X is brand safe.

“I would say we’re certainly using a lot of GPU server capacity,” she says. Of course, “Nvidia’s a great partner,” she caveats.

To Musk: “There are still people within Twitter who care about the platform, who care about making thoughtful, principled, data-backed decisions for Twitter’s users. Listen to them, give them space, don’t overrule them.”

“It’s a chilling effect,” he said. Even with parts of the original ruling overturned, the communications haven’t started up again, he said. “The strategy works even when it loses in court.”

He sees a similar chilling effect happening across the tech landscape. “It’s what they’re doing to academics, folks working in government, to platforms themselves,” he says of people trying to stop content moderation.

Shortly afterward, his address was published online. “I had to sell my house. I had to move,” Roth said. “I bounced between a couple of different places for a few months and then lived in a temporary apartment for a while while I tried to figure out where to land next.”

He expected there to be constraints on what Twitter could strip away in terms of safety. But the company didn’t “behave rationally.” And so things have gotten worse.

Roth is a surprise addition — from what I can tell, Swisher added him into the schedule in the last 24 hours. Quick reminder: she doesn’t run this conference any more... but, you know, everyone still listens to Swisher.

If they’re not ready to handle something, they need to make sure their AI systems can’t do it. Cotra says what she likes about this method is it can “make you pause as long as you need to” to get AI safety right.