Apple’s finally admitting what many in the AI trenches have known for a while: synthetic data can only take you so far. After years of pretending its fake training data was enough to fuel meaningful AI progress, Apple’s now rolling out a new system that actually looks at real-world data—your data. But, in typical Apple fashion, they’re doing it with a giant asterisk: your privacy is still safe. Allegedly.
Here’s the gist. Apple’s working to improve its large language models (LLMs) by using snippets of real user emails—not to be uploaded or stored, but analyzed locally on devices. Only users who’ve opted in to send Device Analytics data will be part of this experiment. Your actual emails don’t leave your iPhone. Instead, your device quietly compares Apple’s synthetic email outputs against your real ones and sends back a single signal: “this one was the closest match.” That’s it. No full messages. No Big Brother moment—at least, according to Apple.
Why all this? Because Apple’s much-hyped “Apple Intelligence” platform has been underdelivering. Those Writing Tools meant to polish your thoughts? Still rough. Summary features in notifications and messages? Hit-or-miss. And Siri—once the crown jewel of virtual assistants—is looking more like a relic from the iPhone 4 era compared to Google’s Gemini or Samsung’s Galaxy AI.
In fact, the Siri situation’s been so “ugly and embarrassing” (Apple’s own words in internal meetings, not ours), it triggered a rare corporate shakeup. John Giannandrea—ex-Google AI head and once Apple’s top AI exec—got benched. The new quarterback? Mike Rockwell, the guy behind Vision Pro. He now reports to Craig Federighi and is tasked with turning Siri from a glorified timer into something that doesn’t get dunked on daily by Alexa.
Apple’s promising big changes to Siri in late 2025 with iOS 19. And yeah, we’ve heard that tune before. But this time, it seems they’re feeling the heat—not just from rivals, but from inside the spaceship in Cupertino.
Here’s what matters: Apple’s finally realizing that great AI needs great data. But unlike its competitors, it’s still playing the privacy-first card hard. Whether that’ll be enough to catch up—or leap ahead—remains to be seen. For now, we’ll watch, wait, and keep our nonsense radar on high alert.
