[seanos]

Assuming the learning algorithm is perfect and always gives the desired temperature, then a saving of 50% means that your desired temperature pattern is 50% cheaper than the temperature pattern you would program with a regular thermostat. That could be true, or it could even be more expensive.

Also, it might be annoying to have a thermostat trying to second guess what temperature you want through a "learning" algorithm, rather than the deterministic temperature you have programmed or adjusted manually.

[bgarbiak]

If the house is cooling down while you are away - don't you need more energy to warm it up again than you would use to just sustain the temperature? I'd say that an 'intelligent' thermostat has greater chance to balance this right than anyone could with a manual adjustments.

[bbest86]

No, I don't think so. Go to the extreme and think about leaving for a week or a month - it is quite clearly more efficient to let it cool and then warm it back up again. On a shorter time scale, your house will lose heat at a faster rate when the temperature differential to the outside is higher, so letting the house cool will reduce the rate of energy loss over the time period the house is cooler. The amount of time you are able to leave the heating off while the house cools will compensate for the higher load when you need to heat it up again.

[asmithmd1]

No, there is no "stiction" with thermal loss. It is most efficient to not add any energy (heating or cooling) when you are not at home. That is as long as you do not mind it being too hot or cold when you first get back home.

[Maakuth]

My guess is that most of the savings come from the auto-away feature that's apparently backed by some sensors (motion detection?). I don't doubt that auto-away could save energy as most people probably don't either care or remember to adjust their thermostats when leaving home.

I don't think that the auto-adjustment really is that much more annoying than programming a regular dumb thermostat. Of course the end result being annoying depends on the algorithm design more than anything else.

[glenra]

They sense three things: (1) light, (2) motion (close-range) (3) motion (wide field/longer range). Presumably if nobody turns on the lights at night or walks past in a while, it decides nobody is home and gradually lets the temperature drift closer to the "nobody is home" range. Then brings it right back up as soon as it senses activity.

It makes sense to start out selling this as just a thermostat, but I wonder how long it'll be before they add a "burglar alarm" module?

[pyoung]

I don't know anything about their algorithm, but I highly doubt that it 'guesses' the desired temperature. It looks like it simply records the users preferences based on time of day and day of week, and then tries to mimic those preferences. It will be interesting to see if they also provide a web based interface that allows users to have more direct control over the algorithm (for example hard coding certain scheduled times for the thermostat to turn on or off)

[seanos]

Not a random guess but a prediction. They do provide a web interface.

[teaspoon]

"Second guess" is UK/US English expression meaning "to attempt to predict or anticipate", so I think you and the parent have the same understanding.