Hacker News Reader

Comment by lawrenceyan | Hacker News Reader

Amazon is a competently run technology company with smart people from my limited experience. I wonder why Alexa is so much worse compared to Google Assistant then.

Perhaps natural language processing / understanding applications aren’t as well researched or developed at Amazon? Maybe it’s the difference in culture between the two companies with Google being much more academic in nature and research oriented compared to the more process oriented operations / logistics focus of Amazon that has caused such a widening gap to build up?

View on HN

Seems pretty simple to me, Google has more data and users that keep adding even more data to optimize their offerings. I doubt Alexa has even 1/10th the users as Google's assistant.

This is the unfortunate barrier to entry when it involves big data. Even behemoths Amazon and Microsoft are stuck with smaller pools of voice data than Apple or Google. It makes it very difficult to compete...imagine being an even smaller player, you have no shot. If Amazon and Microsoft weren't cloud competitors I bet they'd do a joint venture. They are already playing the most friendly with each other.

Amazon has still sold millions of these devices and thus has inordinate amounts of data. When you're talking about training sets that are, say, 1 trillion large vs 10 trillion, does it really make that much of a difference? It's diminishing returns. 10X the data size comes nowhere close to a 10X difference in quality at that scale.

I neither work on Alexa nor on Google Assistant. But based on my experience, going from 99% accuracy to 99.9% would require much more data than 0.9%. Same with adding each subsequent "9".

And the perceived quality difference between 99% accuracy and 99.99% accuracy is much much bigger.

In my experience, like with high availability, each additional 9 requires between double to 10x the resources to achieve.

I think the answer is simply Google the search engine. Google has a huge corpus of semantic data that it gathered from both structured and unstructured data from around the web. Think about how many times you search Google and get a One Box result.

However, I have to disagree to a certain extent. Having both Google (on my phone) and Alexa in my house, overall I have been extremely happy with Alexa. I think it being "so much worse" is very subjective.

This is definitely part of it. Alexa only answers my most basic questions. Google answered me a fact that it was able to parse out from youtube and then pre-labeling the answer with "According to youtube..." That was mind blowing.

A friend of mine who worked at Amazon, not in an Alexa related division so take this with a grain of salt, said that Alexa is basically a huge series of if statements.

I imagine they're working on improving it to more powerful natural langue processing, but it wouldn't surprise me if decisions they made to get it to market years ago are preventing it from being as good as Google right now.

Take what I say with another grain of salt, but I work at Amazon and have a large number of friends that moved over to the Alexa org.

It's not a huge series of if statements. Maybe it was at one time, but 10k developers can do better than that.

I would be surprised if 10k developers could do much of anything. The communication overhead would be enormous.

Amazon is not a perfect company, but they have done a very good job at figuring out how to organize large scale software organizations. Emphasis on SOA whenever possible, teams that aren't too large, and principle engineers who focus on large scale organizational problems.

There's a reason most really nice things are designed by one person.

Amazon is really not that good from a tech standpoint. They are good at selling and shipping at scale. Tech (including AWS) has always been a secondary or tertiary concern for them.

I think this is where a company gets an advantage by doing real research.

That is, Microsoft and Google famously hire a lot of ex-accadmics and let them do basic research not immediately related to a product. That lets them build up 'secret' technology within the company that other people just don't have. However this is an expensive thing to do.

I think this is because google (and microsoft) are companies with a big cash cow and an unclear future path. They need to find and dominate the next big thing in order to one day replace their current thing.

Amazon on the other hand has a number of avenues for future growth. They just need to eat all of retail and internet infrastructure. Therefore they can grow just by solving their customers problems in a focussed way. Low cost, well engineered, current but not next generation stuff.

Google had all of the pieces before Alexa even existed. The Google Assistant was basically a bunch of existing Search tech duct taped together at the start.

Some of this investment is negated by the amazing progress in neural nets since Google was founded, and diffusion of Search knowledge, but not all of it.

  > competently run technology company

Not all parts. Not even remotely.

Source: worked there.

You’re at Google now right? You people think anybody not at Google is an idiot.

The reality is Amazon is extremely competently run given its size. The pace of innovation is astonishing. We might not be ubermensch but we get the job done, and we are trusted by more people in this country than either Google or Facebook are.

Your at Amazon, right? You should be aware that many other Seattleites in tech despise those who are part of FAANG, vast data mining (FANG) & extreme internal secrecy (Apple, Amazon) have created an environment where ethics are ignored, and only bootlickers are promoted.

This has come back to haunt some of these companies already, but fixing the structural rot is unlikely to happen anytime soon.

Alexa is new type of interface, just doing better AI/voice recognition is not enough. Amazon never were good at User Interfaces. Google is getting better at it I think, has more designers and UX.

Amazon very much has a "get it done" culture. Get a bunch of good engineers in a room to solve a very specific problem. They've never been really focused on iterating fuzzy problems.

Reply on news.ycombinator.com