Alibaba has a voice assistant better than Google’s | Hacker News Reader

Alibaba has a voice assistant better than Google’s

technologyreview.com

59 pointspalad1n36 comments

So, wait, this conclusion was reached because of a single pre-recorded demo call, compared I guess, to Google I/O's prerecorded demo calls?

If it's servicing millions of live requests and is in production, why couldn't they review the live system?

The story's title kind of seems clickbaity because of this IMHO.

I just hate that this web-site demands to exit incognito mode or log in in order to read the article.

Works for me in Safari. What browser are you using?

Works in firefox as well, but in chrome:

https://i.imgur.com/KTN1Jnf.png

Perhaps turn off Javascript? I use a nice little JS toggle plugin in Firefox for that, and apparently I disabled JS for this site in the past, which is a sign they tried some BS when I last visited (and the site functioned fine without JS) :P

Even worse, on Chrome mobile, the bottom section is cut off and unscrollable.

I'm guessing it's to stop the soft paywall workaround of opening in incognito?

I'm using Brave

Brave is based on Chromium, how are they able to claim "loads major news sites 2 to 8 times faster than Chrome/Safari on mobile and 2 times faster than Chrome on desktop"?

Just simply by running an adblocker?

That's a good question...I don't care about the performance of a browser because nowadays everything loads instantly anyways. Brave has built in anti-ad features, also can operate in TOR mode and that's pretty much the only reason of why I'm using it.

I hate it, too. In any case, if the incognito mode can be detected, it's a bug in the browser. Way to go, Google.

Slightly surprising that Google Chrome is bringing up the back of the pack when it comes to incognito mode, Firefox and Safari can both view this site without the popup complaining about using incognito mode.

Not defending Chrome, but if websites are willingly blocking users carrying about their privacy, then maybe you shouldn't read those websites.

Paying with your wallet or your eyeballs works. Most of the time reading such articles is a time waste anyway.

One doesn't exclude the other. I'm going to let the website owner know, but at the same time I'd expect the browser vendors to take greater care about protecting the private mode.

Maybe the site developers simply didn't care about detecting the mode in other browsers.

> Slightly surprising that Google Chrome is bringing up the back of the pack

Actually that's quite opposite to surprising in this case - tracking users and their behaviors is googles bread and butter so this is probably feature, not a bug

Apparently private browsing is pretty easy to detect in any browser:

> Also yes, in case you didn't know, there's lots of ways to detect private browsing using JavaScript. It's kinda dumb that browsers don't work harder to make it imperceptible.

https://toot.cafe/@nolan/101189400134875073

It's "servicing millions of customer requests a day" or some part of "50,000" (later in the article)?

The example doesn't seem compelling to me, it's not rearranged the delivery, it's not solving any substantial problem in the co-dialog. What does this prove?

It's possible that it's actually serving millions of requests, if most requests happen outside of customer service calls or if they count each individual utterance in a dialog as a separate request (it probably is a separate request to the backend).

On the other hand, maybe someone wrote a description in Chinese that said "several tens of thousands" and someone else had it machine-translated into English. The problem being that "tens of thousands" can also be a symbolic "big number" (like "myriads") and modern English tends to use "millions" as that symbolic number instead.

The existence of symbolic numbers is a common source of translation mistakes. For example, I recently came across a Japanese text [45] where Google Translate will turn 一三 into 13 or a thousand, depending on whether you include 「quotation marks」 or not

[45] Footnote 45 of https://www.aozora.gr.jp/cards/001850/files/57353_57270.html

https://translate.google.com/#view=home&op=translate&sl=auto...

I always wondered what the implications are for the "lack of privacy" in China, for the better or the worse. What I mean by this is something like WeChat. A lot of Americans do not understand the sheer size of data it collects. WeChat (by Tencent) is pretty much a conglomerate of all the top services we use here in the states (Google, Paypal, Facebook, Stripe, Venmo, Whatsapp, etc). Now imagine when the Chinese government starts dictating the development of AI through this data. I am not sure if there is another country that can collect and distribute data at this scale since there is not much public remorse for such actions.

This is why China will beat any other countries including US on AI and ML hands-down, it has no privacy law and people don't care that much either so it collects such a huge volume of data for training and mining.

Wouldn't that be why it wouldn't. It can't export their models because they require to much information. In places like europe it would be a nightmare to use any of it. So in China it would be ok, but outside it's hard to implement.

So which phone number can I call to try this out?

Considering many Alibaba’s P8+ comes from FLAANG and the whole company works like hell (9-9-6) to deliver, this is not surprising. Serving millions requests is laughable at Alibaba scale and might indicate this is not widely used yet.

And, what so???? I can't understand the reason for such fascination.

Because Google is popularly assumed in the West to be more technologically advanced.

Not all languages are created equal. Chinese has very few sibilant consonants. It all about vowels with varying pitch. English is just hopeless. I myself for example just cannot use Google in English, because of 50% error rate.

Isn't phonetics already a solved problem in ASR? The hard part is NLU, including semantics and pragmatics, which this article mainly talks about.

Some languages are significantly easier to train against than English, for example 80hrs of Turkish got some decent results for this group using Mozilla Deepspeech: https://arxiv.org/pdf/1807.00868.pdf

For context, Baidu used a bit over 5000 hours of English to get a decent model.

Also, OP may have a pronunciation difference (regional, hyper-local or just from learning English as a second language) that isn't handled well by Google's models.

I suspect deep learning researchers have no real incentive to use the minimum amount of data sufficient to get their thing working. The worst case for them is if their model underperforms because of insufficient data instead of something truly DL-related, so they use as much data as they have/can handle. It could still be that you're right, but I wouldn't rely on those kinds of published numbers.

Whatever, but it does not work, as I can easily demonstrate. I once tried to buy flashlight batteries from Vancouver. I tried every possible variation of BATTERY, but had to write it on paper.

Problem is that English does not have standard pronunciation, because writing is not phonetic. In other languages everybody knows what is the official pronunciation even if their own dialect vary wildly.

Just tried saying battery in many different ways, even mumbling and it got it correctly every time. No idea how you pronounce stuff in order to have such problems.

The problem at time was the stress point. In most of my languages the stressing is flat or in the back. At first I started using the standard "BAT-TE-RII" of almost all languages, but realized soon it is maybe pronounced "BAT-RII", but failed to emphasize the first syllabel as in "BATTT-ri" and ended up varying forms of "baaat-RRRIIIIII".

Mandarin Chinese has more sibilants than English, so I don't know what you're talking about. The English sibilants are s, z, ts, ʃ, ʒ, tʃ, dʒ. The Mandarin sibilants are s, ts, tsʰ, ɕ, tɕ, tɕʰ, ʂ, ʈʂ, ʈʂʰ.

Actually English has thousands of sibilants and millions of vowels . Think about differences between a Scottish speaker and Guido van Rossum. Problem is of course that there is no standard pronunciation or even method to define it.

We have IPA though, accents can be defined as certain phonetic replacements. It's not that simple, but it isn't impossible either.

For example this chart: https://en.m.wikipedia.org/wiki/International_Phonetic_Alpha...

Reply on news.ycombinator.com