… OpenAI also said it is expanding its alignment efforts across the training process to prevent “reward hacking,” a behavior in which AI models pursue their goals through unintended or undesirable means. The company says it plans to share more details about this work in the future. …
… When OpenAI finished training o3 in early 2025—a model optimized for coding even more than o1—it finally had the foundation to build a real AI coding product. But Claude Code was already poised to launch publicly. OpenAI was scrambling to stand up a competing product. …
… Following incidents of high-profile rogue activity by some of its AI agents, OpenAI said this week that it is halting model training runs and overhauling internal safety protocols. The company said that its upcoming Astra model may represent a turning point of “critical” cyber capabilities. …
… Employee training and a paid monitoring service have helped FLS fend off most shady carriers. …
… I think there was fear that Google would be training its models on the A24 catalog and that's specifically not part of the deal. …
… One of the reporters who won our first Pulitzer that involved AI-supported reporting, what he said to me is, "You have to think of AI as a filter.” In his case, he was examining bomb craters in the Israel-Gaza conflict, and he said that it regularly makes a certain kind of mistake. …