What the "expert" has done here is taken the worst features of multiple meters, and put them together as though every meter is this way. And even then, he's overstating things...this "they can tell if you're home by how much electricity you're using!" bullshit has been around forever, and it's ridiculous.
Let's see, where to start. One, almost no meters use GSM. GSM is expensive on a per-device basis (the target upper limit for hardware costs is about $100/meter), poorly-supported by cellular providers...with future-state being no support at all...and renders the utility dependent upon an outside provider for all of their network backhaul from the meters. This is why, if you look at any of the major meter manufacturers (Itron, Elster, Landis + Gyr, etc.) you will find that they all use a very different architecture that does not at all rely on GSM, or any other cellular protocol. They use mesh networking and collectors.
Second...okay, let's talk about what you can do with the meters. Yes, theoretically (it's never been done), you can figure out if someone is home. You would need to be in their neighborhood to begin with since you have to speak directly with the meter. You would need to reverse engineer their specific approach to frequency agility, and break the crypto so that you could then impersonate the head-end and do meter data requests. With that, you could do data sampling to determine what normal peak and low usage numbers were, and from that you could derive whether or not they were probably home at any point in time. Or...you could simply walk near the house and see if the lights were on or there were less cars in the driveway/garage than usual. Which thieves already do, as a best practice that works pretty well.
Then, let's talk this "house fire" over "overload" bullshit. Meters do not regulate power. Let me say that again. METERS DO NOT REGULATE POWER. They can turn power on and off, and that is it. They cannot modulate voltage, wattage, frequency, or amperage. And while in the early days of AMI adoption it was feared that a compromised head-end (or impersonation thereof) could permit an attacker to issue enough remote disconnects to cause what's known as a "bulk load shedding event," it turns out that the meters and their communications networks are too slow. That network architecture I described above with collectors and mesh networks? Every approach in broad use acts as an inherent throttle on communications in bulk. So you couldn't even destabilize the grid; the effect would happen too slowly. And just as the attacker could turn the meters off, the utility could just turn them back on..so this would not be what you would consider a "blackout." They cleverly cite a house fire, though that was the result of a meter vendor changing the polymer used in the meter backing; the replacement polymer had the dual properties of 1, not being ablative (so it could catch fire) and 2, being more brittle...so if the meter wasn't seated the right way, it would crack. An arc would form eventually, setting the meter's base on fire...and there's your house fire. Nothing to do with hacking in the least.
This guy Rubin is a wanna-be with a new company, and he's decided to look at devices which are widely used without really learning about the industry they belong to, or getting the experience needed to know how all of this stuff really works in detail. He's not a widely-recognized "expert" in cyber security, neither in general nor within the power industry.