It's all good, it's all good' says Mark Zuckerberg (www.pcgamer.com)

At Meta Connect 2025 Mark Zuckerberg unveiled the second‑generation Ray‑Ban Meta smart glasses — billed as a Live AI, multimodal assistant paired with a muscle‑controlled wristband — but the onstage demos repeatedly misfired. During a cooking demo the glasses listed ingredients instead of giving step‑by‑step help, then suggested nonsensical actions (like grating a pear into a nonexistent sauce) after long pauses. Later, Zuckerberg couldn’t accept an incoming video call using the wristband gestures: the UI showed the call but the device failed to act, leaving an awkward, persistent ringtone in the hall. For the AI/ML community this is a useful stress test of where consumer multimodal devices still struggle: real‑world robustness, latency, sensor fusion and grounding, and human–machine interaction. Failures point to likely weak links — network dependence or cloud latency, unreliable vision‑to‑language grounding (object recognition vs actionable instructions), speech/intent recognition timeouts, and brittle gesture decoding — all under the glare of live conditions. Technically, it underscores the tradeoffs between edge compute and cloud services, the need for offline fallbacks, rigorous real‑world testing, and UX fail‑safes. While the publicity keeps the product in the conversation, these demos highlight that impressive research prototypes still require substantial engineering to be dependable in everyday use.