I.
I’m exploring the Faroe Islands this week! Think charming villages and breathtaking scenery; a bit like Iceland, but with fewer people; Scandinavian waffles and so many rainbows. I’m posting plenty of photos and stories on Instagram.
II.
404 Media reported on OpenAI’s practice of hiring contractors to read “a massive stream of real users’ ChatGPT prompts … to improve the responses ChatGPT gives to its users.” The problem is that prompts often include sensitive personal information, despite the company’s efforts to anonymize them. I’d argue most people likely don’t consider that other humans can read their chats. OpenAI’s “improve the model for everyone” setting is enabled by default, but you can turn it off. (Anthropic does this too, by the way.)
III.
I spoke to Alfred Ng at Politico about AI, privacy, and the need for user awareness and control. Just like Anthropic and OpenAI can do more to help people understand how model improvement works, Meta can do more to help people understand where its AI gets information from.
IV.
I spoke to first year journalism students at New York University last week. In addition to practical security steps and threat modeling, we also discussed the FBI’s January raid on a Washington Post reporter’s home; the Trump administration attempting to subpoena three New York Times reporters in June; and WhatsApp’s Strict Account Settings for enhanced security.
V.
Ethel and Julius Rosenberg were convicted and executed by the U.S. government in 1953 for passing atomic bomb secrets to the Soviet Union during the Cold War. The couple left behind two young sons, ages 6 and 10. A longread in The Atlantic tells the story of Robby and Michael, now 79 and 83, who are still trying to make sense of what happened.