3 ms·

It doesn't even matter how sincere they are. False negatives hurt the perception of the product, because then people need to repeat the key word. False positi

It doesn't even matter how sincere they are. False negatives hurt the perception of the product, because then people need to repeat the key word. False positives do not hurt the perception of the product, because the recording occurs silently. Their incentives are aligned against privacy, and so privacy will suffer.

vidarh 7y ago
False positives does hurt the perception, because the device will usually do something or utter a response unless it fails to detect something intelligible or writes the request off as not intended for it (having two devices at opposite ends of a large open space living area can sometimes get frustrating, but looking at the history, you see how often they correctly determine which device should respond) An occasional false positive is funny, but it gets old pretty quickly.
- jackpirate 7y ago
  That's not true. Amazon can have a significantly more accurate classifier running on their backend that processes the transmitted audio. And this classifier can be more accurate due to both more computation and it would have access to the entire data stream, not just the initialization words. You would never know that your data was sent to Amazon and then ignored.
  - vidarh 7y ago
    They can, and they do. You can see that in your history because you can listen to the audio of things it decided subsequently was not intended for it. But unless you have multiple devices, those make up a relatively small proportion of requests, and hence most of the time false positives is something they have a strong incentive to stop because it causes negative user experiences. [If you have multiple devices it will happen "all the time" when a request is genuinely meant for a Alexa device but more than one device hears the request, but then it has no UI effect - one device will usually answer, and the other(s) will stand down; this is by no means perfect, but it works reasonably well]

vidarh 7y ago

False positives does hurt the perception, because the device will usually do something or utter a response unless it fails to detect something intelligible or writes the request off as not intended for it (having two devices at opposite ends of a large open space living area can sometimes get frustrating, but looking at the history, you see how often they correctly determine which device should respond) An occasional false positive is funny, but it gets old pretty quickly.

jackpirate 7y ago
That's not true. Amazon can have a significantly more accurate classifier running on their backend that processes the transmitted audio. And this classifier can be more accurate due to both more computation and it would have access to the entire data stream, not just the initialization words. You would never know that your data was sent to Amazon and then ignored.
- vidarh 7y ago
  They can, and they do. You can see that in your history because you can listen to the audio of things it decided subsequently was not intended for it. But unless you have multiple devices, those make up a relatively small proportion of requests, and hence most of the time false positives is something they have a strong incentive to stop because it causes negative user experiences. [If you have multiple devices it will happen "all the time" when a request is genuinely meant for a Alexa device but more than one device hears the request, but then it has no UI effect - one device will usually answer, and the other(s) will stand down; this is by no means perfect, but it works reasonably well]