Hacker News Reader

Comment by sfblah | Hacker News Reader

Siri's awfulness really is a thing to behold. I haven't used an android phone in a while. For those users out there, does its voice assistant actually work?

View on HN

The current situation on Google's Android Pixel phones is odd. The old non-LLM Google Assistant works well in a limited domain: Things like setting alarms, phoning by name, etc. It's similar in scope to Siri, but with better voice recognition, and better context awareness. However, Google is desperate to kill Google Assistant and force all Pixel users to use Gemini instead. Gemini 3 is a very good LLM, and far, far, far more versatile than Google Assistant. But Gemini won't do the simple things as reliably as the old Assistant. Setting an alarm works maybe 90% of the time with Gemini. If you asked the old Assistant "What time is it?" it would respond "It's 4:40 PM". If you ask Gemini "What time is it?" it will sometimes respond "It's 4:40 PM CDT in {your city}", but sometimes it will say "It's four four zero Pee Em in {your city}" and sometimes it will do a web search. Results are spotty in other areas like voice dialing. I've retained the old Assistant, because I want to do the basic things far more often than I want to verbally vibe code. But rumor has it Google is going to disable the old Assistant in March, forcing all users onto Gemini for voice commands. Unless Gemini gets much better at handling simple tasks by then, Pixel users will end up with a voice assistant much more frustrating than Siri.

Not to mention all the useless LLM fluff Gemini has. I turned off Gemini because simple questions like "What's 1 USD in AUD" would be met with 30 seconds of "As a large language model, I can't provide investment advice, so it's always important to check information yourself [...], but one Australian dollar is approximately $0.65" (note the conversion in the wrong direction). By comparison, Google Assistant just gives you a number straight away.

They should have deterministic actions for certain non fungibles and let the LLM do the rest.

Its gemini so at least its smartish and has some integration with the rest of the ecosystem so it can do some assistant work as long as its read mostly, but integration with the rest of the phone is almost non existent. It also struggle in noisy environments and in mixed language situations

No. actually no company on earth has solved the voice assistant thing yet

How is that even possible with modern transcribing and natural language capabilities?

Exactly. Looks like everybody's complaining that Siri isn't a better Ask Jeeves, when that's not the design goal. What people expect is an LLM that has full access to the phone. Nobody's even remotely close to shipping that.

I just want "set timer for 5 minutes" and "what time is it" to work reliably, 100% of the time.

Works on Siri! You can even set – get this – multiple timers ;)

I seem to get about 90%; wife gets closer to 80%. I just want it to be better.

Activate Siri and then say the amount of time you'd like on the timer. For me, that's holding the side button then only saying, "20 minutes."

The one caveat is that once the timer you want is two and a half hours or longer, then Siri replies by asking what you would like to convert to.

That may be it - we’re always asking a HomePod to do it - “hey siri set timer for five minutes”

My Siri-initiated timers are always done with my phone, probably 50 or more each week (work stuff). The only time I get a failure is when I release the side button too quickly. I've made certain the spoken feedback is enabled to reduce the risk of me making that mistake. (Settings > Siri > Siri Responses > Prefer Spoken Responses)

As for, "What time is it?"... Try activating Siri and only saying, "Time."

I suspect that's the main difference; if you're trying to use hands-free voice activation via "hey Siri" you get a much different experience than if you can touch the watch/phone to trigger Siri first.

And thinking back over it, more than half the failures are complete - e.g., it likely never activated at all. Very few are "it set a timer, but for the wrong time".

Good chance that's what captures our different Siri experiences. The few times I've done it spoken was always with AirPods and I always waited for the Siri reply (been a while; is it, "Uh-huh"?) after I said, "Hey, Siri." But my experience activating Siri with speech is so minimal as to be untrustworthy of anything broader.

I guess it depends on what you mean by voice assistant.

But Android has been 100% accurate for simple commands for over a decade for me. Things like:

- "weather." tells me forecast of where I am.

- "alarm at 8am" and "alarm in 30 minutes" works as expected.

- calendar commands also work

My favourite is "go home" which opens Google map with a route to home.

These things just work. I don't recall last time I had to repeat myself.

I wonder if we are getting different versions based on geolocation (I'm in Europe) because my experience is the absolute opposite of this. I actually had the thought "maybe I should switch to apple to stop having to deal with this" just this week (although reading this thread siri is as bad).

My experience is only through android auto and it honestly makes me furious how bad it is. There is absolutely no other tech product in my life that gets even close to how bad voice commands are handled in Android.

In my experience, literally everything sucks:

- single language voice recognition (me speaking in English with an accent)

- multi language voice recognition (english commands that include localised names from the country I'm in)

- action in context (understand what I'm actually asking it to do)

- supported actions (what it can actually do)

Some practical examples from just this week:

- I had to repeat 3 times that "no I don't want to reply" because I made the mistake of getting google to read a whatsapp message while driving, and it got stuck into the "would you like to reply" (it almost always gets stuck - it's my goto example to show people how bad it is)

- I asked it to queue a very specific playlist on Spotify, and it just couldn't get it right (no matter how specific my command was, I couldn't get it to play a playlist from MY. account instead of playing an unrelated public playlist)

- I asked to add a song to a playlist, and it said it couldn't do that (at least it understood what I was asking? maybe)

And in general I gave up trying to use google maps through voice commands, because it's just not capable of understanding an English command if it contains a street/location name pronounced in the local language/accent.

If they didn’t have so many limitations imposed for safety or by permissioning. Hey Google call my wife. 30 seconds later “something has gone wrong”. Or hey Google play _______ on YouTube music. “Playing something else on you tube music”. Its stupid.

I use it almost daily, to set timers and alarms

Same thing I used it for 10 years ago hahaha

I used to use it for that. A few months ago I got an Android system update and it no longer works for that. It just does web searches if I try. Now it's trying to push me into this thing where it takes a screenshot and tells me what's on the screen. I've never once cared about that.

Failing to find any way to get the alarm thing back, I turned off the entire assistant thing.

It works okay. I like that it's universal (the same assistant on my phone, on my home devices, in my car, in my earbuds). I like that it does tasks right, but you have to know how to phrase them (my most common is probably "remind me to X tomorrow at X time"). Setting alarms and timers, creating calendar events, asking about the weather on a specific day or in a specific place, asking how long it'll take to walk/drive somewhere -- all good. But anything more complicated than that and you get erratic behaviour. From what I've seen with my friends interacting with Siri, I'd say they're about equal in capability.

It works pretty well for me, but doesn't do nearly what I'd expect.

EG I can talk to it like I would chatgpt and it works well. But I can't be like "hey I want to get dinner with my wife on our anniversary, please book the best available option in my city for fine dining"

It's still way better than Siri, which feels like a voice CLI to me (same as Alexa, which is very low quality IME)

I don’t think I’d want to talk to a voice assistant like that. Maybe it’s a generational thing? Things like that are ambiguous enough discussing them with human beings and a big part of things like voice assistants is understanding how it’s going to interpret and execute a response based on what I say to it.

Pretty much the same as Siri. Alexa isn't any better either.

Could you share a bit about your use case/experience? Siri does what I need it to do— send messages, create reminders and calendar entries, look up basic facts and cites the source, play music, add things to lists, etc. I’m curious if you’re trying to do things that I haven’t, or if you’re just having a very different experience with those same things? Or maybe just have higher expectations for it?

Edit: why in gods name are people downvoting me for politely asking about someone’s differing experience?

The very first thing I tried when Siri was released -- "set an alarm for ten minutes before sunset" -- still doesn't work. "What time is sunset?" and "Set an alarm for 5:03 PM" both worked on day one, and still work. Zero progress.

Interesting. I think I probably mentally separate the information retrieval realm and the command execution realm more than makes sense for the interface. There’s no apparent reason that shouldn’t work based on what the user is given.

I have never been able to play music with Siri. Even something as simple as “play this specific album” gets completely bungled.

Interesting— I use that functionality constantly and listen to a wide variety of artists, some of them pretty obscure. Do you use Apple Music or another service?

I can think of one time recently where no matter how I prompted it to play an album for (decades old but probably triple platinum,) it kept playing some cardi b song with the band’s name in the title instead… but that’s probably like a 1 in 2000 request problem. Maybe its a genre thing?

It’s extremely hit or miss for me. Sometimes it works and I’m amazed. Other times it fails to play my main playlist that I’ve played 1000 times before.

Huh. I’d love to see what their UX testing looked like. It clearly missed someone doing what you’re doing with it, sadly.

Yea, I’m boggled. At this point Siri should be able to parse and understand a wide variety of forms of the same command, but it still seems to fail. This should be doable even without LLMs.

Yeah I can ask it to play specific editions of specific EPs named the same or similar thing to the albums or whatever and it rarely screws up. There’s got to be something significantly different in our approach. I wish I could test it.

"Siri, play The Dragonborn comes at 25% volume"

"Here is what I found about "The Dragonborn comes at 25" on the Internet" opens Safari

Ah I never felt inspired to use it on a computer and always use physical volume controls in the car and through headphones, so I wouldn’t have run into that. It does seem like something that should be a day-one sort of feature.

"First you have to unlock your iphone"

(╯°□°）╯︵ ┻━┻

I swear I've gotten that response, from a nearby HomePod, while holding my unlocked iPhone in my hands.

I don't know what it is doing sometimes.

The only time I find Siri useful, or I should say ~potentially~ useful, is while driving text, call and to ask basic facts. The amount of times I've heard "I can't show you that right now" after basic questions is insane. I just stopped asking it questions. Recently I asked "what engine is in a 2022 f150". Trying it without Carplay now, it literally just displays text. It should be able to TTS those results. What on earth have they been working on if not things like that?

I know there at least used to be a setting to specify if you get a verbal or text response based on whether or not the phone is locked. Maybe that would get it to stop just displaying text?

I pretty much only use it when I can’t look at the phone so I’m not sure if it’s still there.

> Siri does what I need it to do

Not true for me at all, it fails at the most basic tasks, sometimes even at tasks it has done before. Three examples:

- "Timer 5 minutes" -> Loading spinner is shown. Siri disappears after a few seconds. No error, no confirmation. I then have to manually check if the timer was set or not (it was not).

- "Turn on the lights in the living room" to which it responds "Sorry, I cannot do that". I have Phillips Hue lights that are connected to Apple Home, of course Siri can do that. It did that before.

- "Add tooth paste to my shopping list". The shopping list is a list I have in reminders. It then tries to search for the query on Google. I then tried "Add tooth paste to the list shopping list in reminders" which worked, but if I have to be this wordy, it is no longer any convenient.

There are many more simple cases in which Siri always / sometimes fails. I also have the feeling that it performs far worse if asked in my native language (German) than in English.

Yeah that’s strange. I set timers constantly both at home and work and I can’t recall a single time it hasn’t worked. I periodically add things to lists without issue. I have zero experience using it in another language. Maybe their testing sucked for that?

The lights thing is insane and annoying.

"Turn on living room lights" - "There are no lights in the living room"

"Turn on living room lights" - lights turn on.

Naming some of my lights Greg has helped a bit.

‘Siri turn on torch’. Used to work, now all I get is “sorry, Torch isn’t available right now” this is at night when it is plugged in and I need to work as a nightlight to go open the bedroom door to let the dog in or out without blasting myself awake with the main phaser array next to my bed.

Reply on news.ycombinator.com