I don’t see where a government would need a chatbot. Anyways, chances are that half the staff was already using some form of LLM before this trial.

The article you cite states that accuracy drops to 60% if the enrollment and testing data were collected at different sessions. I imagine the effects of coffee or walking on heart rate would make that even worse.

Your neighbors WIFI signals are too weak to matter in this case. Even if they were strong enough, this is a receiver-transmitter setup, so it would still be impossible to do unless you connect to their network. Even then, they’d have to assume you’re the only person present between the transmitter and the receiver.

Presence detection through WIFI was already garbage enough, this one is plain unusable.