Ask HN: It's 2024, why are voice assistants still so buggy?
Still feels like 20% of the time I try to use Siri/Alexa, they hit one of these failure modes:
1. Fail to activate
2. Fail to respond (just keep listening forever)
3. Activate on the wrong device (talking to Siri on Phone, HomePod 2 rooms over responds)
4. Fail to access data they should have ("Who is speaking?" or "I'm having trouble...")
Thinking over it, I think voice assistants have the most bugs of any technology I use daily. It's because they're a cost center. They don't generate revenue for any big tech company. No one makes purchase decisions based on the voice assistant quality. I know a few people who lightly use them for timers or music, but almost no one goes deeper than that. The technology as such isn't bad. It's just the idea that the gear is spying on me, and I don't know where the data go for processing, that is the big downer. It’s not that. Otherwise no one would use gmail. The difference is that gmail works Yeah, makes sense (but still disheartening). I wonder if Apple will roll out a paid Siri Pro tier (with LLM). Maybe that would incentivize investment into tablestake features (ie. responding reliably). I think Amazon is the main one being talked about. Alexa was created with the thought that people would use it to order more stuff, they didn’t. They even gave away the Amazon Dash for free, thinking it would be the Alexa for the kitchen that would get people to order food, it had a barcode scanner. Amazon items and prices are too inconsistent for anyone to order anything with a voice assistant and they never solved for this. They’d need to end the entire 3rd party seller program to even have a shot. Siri is more of a value add for people buying Apple hardware, like everything else in iOS and macOS. It can also be seen as an accessibility feature. It was never really meant to be its own revenue stream. It’s funded by hardware sales. Actually apple should have most incentive to make it reliable, as they sell least amount of other stuff based on ads and recommendations, afaik. They should, but Siri is astonishingly, impossibly bad. It's the worst of the bunch. I guess that's evidence that Apple really isn't competing on quality of most of their products. Once you're locked in their ecosystem, you just take what you get. I built an Alexa skill at work. It's mostly a ton of edge cases and prompt variations you have to handle. Alexa also has a very short time out, something like 8 seconds, so your skill will time out very easily. Handling state is a whole other fiasco. Some things are out my control, like if Alexa picks up the skill invocation phrase or parses the response correctly. I suspect LLMs will help with this. Alexa is already integrating LLMs with their newer skill kits. For #3 I recommend setting up your phone to respond to "Siri" but Homepod to respond to "Hey Siri". Because everything is buggy these days. ;-) 1. Scottish accents ...
2. See #1.