Heck, they've been doing machine learning since at least 2004, when they added Latent Semantic Analysis for junk mail filtering to the Mail app. And does anybody remember iTunes Genius (collaborative filtering on your playlists in the cloud)?
I think Apple has traditionally done a really good job of keeping the technology used under wraps, for better or worse.
Did Genius use AI? I thought it was more of a collaborative filtering feature where it used other people's playlists to build relationships between songs.
There's a chapter on Bayesian stuff (maybe more, the book is at home) in "AI a Modern Approach" so I'd say the answer is yes. I remember it vividly because they use a dentist/toothache example :P
I know "it's in AIAMA" is hardly a formal definition but it's good enough for me.
There is nothing intelligent about machine learning. It’s cool but far from intelligent. AI buzzword to describe machin learning is equivalent to people calling computers cpus. Let’s me know really quickly just how clueless they are.
If a calculator is AI then a wrench may as well be AI. There is huge disagreement over what ~really~ is Artificial Intelligence so the argument is pointless without a chosen definition.
Wow, so your comment was even more useless. Or was it a joke that I didn't get? The challenge was "prove me wrong" (with the opinion that I just uttered), not "disprove that this is my opinion".
Fair point, I mentioned it because I was thinking they combined the LSA with some form of classifier that got trained as you marked messages as spam (or told it when it misclassified), I'm not sure what Apple's full implementation was though.
Yup - it seems like they used Adaptive Latent Semantic Analysis as a dimensionality reduction technique to represent knowledge[ 0]. LSA helped them use the filtering technique on the device itself as it was not heavy.
Apple has actually really really fallen behind. Their speech reco is shit. Their maps are super shit. Siri search is so bad and gets in way all the time that I turn it off right away on my iOS devices. Their photo management and image search is literally circa 2000. The spell check and completion is circa 2007 (ie no significant improvement since first enabled). They can’t get right even about when to auto-turn caps. Large number of badly spelled and misunderstood messages in the world can now be attributed to utter incompetence of Apple in AI arena.
To me it wasn’t a surprise that Apple finally got out of self driving car project. They simply don’t have cultural aspects to undertake such a massive AI heavy project including hiring and motivating the necessary talent. This is prime example of how money can’t help if you don’t have right culture. They focused on design so much that ML and AI were relavent only to the extent it helped design. Thing is that such pragmatic view doesn’t work in highly exploratory and rapidly advancing field where 90% of the things have no apparent business use but you do need those in the hope of cross-pollination with other 10%.
Apple is not an AI company and is not setup to be an AI company. There will be massive cultural shift that would be needed to turn this around.
Apple Maps has improved significantly over the past few years. I now have no issues using it for every situation, except for some place where they don’t have transit info. Still need google maps for that.
I don't know who is in a worse position. Apple has the problem of a possible trade war. Google on the other hand is going to be regulated with GDPR and whatever the current administration will have to pass due to Facebook. I think I am leaning toward Google having a worse situation due to this move.
Google has effectively been GDPR compliant or close to it for years. Takeout, view and edit your ads profile, industry-leading internal controls on production data access, etc. They’re a first party, not a data broker, so will have an easier time with “legitimate interest” arguments. They also helped develop the law. GDPR is not a big deal for Google. It’s a big deal for similar organizations that have to suddenly catch up to Google’s security and privacy posture.
I mostly agree with you, but the penalties for non-compliance are so staggeringly large, the requirements are so vague, and the EU's love of fining American tech companies is so great that Id' be awfully nervous about GDPR if I were a large American tech company.
Now that is just being prejudiced. It's not EU's agenda to try and bring down American tech companies for the sake of malice (or protectionism). You do not think they have not been at some fault? Or that there is something good about GDPR too?
While it's easy to start mulling over why and how things are or could have done differently I'd argue that there is no animosity towards American companies per se. The data of its citizens however is a very sensitive issue for EU and in the light of Snowden leaks maybe they are rightfully fearful of US having too much technological hold over them. There is silly people over both sides of the ocean but let's not start willfully feeding rash preconceptions for no reason.
The claim that EU may have a populist-fueled agenda against American tech companies is not entirely without basis. It's not lawless anti-Americanism, but the tone has certainly changed.
Consider what the Economist has to say about Margrethe Vestager, who runs Europe's trustbusting agency.
The Economist criticizes Vestager for fining Google and Apple. It concedes that both were anticompetitive, but complains that there is no proof "that consumers were denied a superior service as a consequence". But that's an impossibly high bar: no one can guarantee that competition will emerge; regulation can only provide a level playing field. So I don't think that this is a good article to prove your point.
I know where those comments come from. It was the NYT mainly and I think Reuters or Bloomberg doing it a bit, too - they were pushing this angle that the EU is doing it out of "jealousy" and nothing else.
Some people have actually bought that idea. But I imagine they'd want the US government to exhibit some of that jealousy against Facebook right about now, too...especially with Facebook now announcing that they won't implement the GDPR globally.
Ironically, it's the same media entities that seem to have essentially admitted they got it wrong with all the negative posts they've published against Facebook recently, and I think those sorts of articles will continue against Google and other privacy-violating companies, too, in the near future.
I've seen constant protectionist sentiments from Europeans, Americans, and everyone else. You are delusional if you think it doesn't exist and it's critical to address it on its face when considering legislation.
You're right, it's not the EU's agenda to try and bring down American tech companies - it is strictly Margrethe Vestager's political ambition to ride the hardline, protectionist stance against American tech straight to the EU President's office. Thankfully, she's burned so many political bridges that her own country is railroading her plans.
One counter example would be Booking.com. It's a Dutch company through and through[0] and pays the vast majority of its taxes in the Netherlands. Yet at least one of the European competition (!) authorities has already tried to use GDPR as a means to try and prove wrongdoing before GDPR is even in effect.
This is to say: you don't have to be an American company to end up in the crosshairs of overzealous and downright populist regulators.
Source: used to work there, had to help respond to a rather ludicrous 40-page list of questions.
[0] Headquarters in Amsterdam, culturally strongly influenced by the Dutch, the biggest concentration of staff and virtually all product development in Amsterdam, albeit owned by an American company.
Quite possibly, but Google and Facebook are known entities that can afford flotillas of lawyers. The next up-and-coming startup to reach the "backlash" stage of the hype cycle will be a much more vulnerable target to European regulators. And if it's too easy for some hated target to comply, then the government can ratchet up the regulations even higher, and Google and Facebook will have the resources to keep up while the newer market entrants will be fucked. Google and Facebook will only be at risk if they deliberately flout regulations the way e.g. Volkswagen did.
European here and I feel this is a bit shortsighted. Not really sure the sentiment is to target US companies (if for nothing else ours are quite open/interconnected economies and there's lots of fondness for US services - check both market share as well as revenues) but really.. Facebook's tax of 5k/year in the UK, the double irish with a dutch sandwich, the backroom deals from the 80s with the Irish govt.. these are global practices which truly devastate everyone in the long run. I actually trust bureaucrats to step in and audit - and subjectively think they very often do a decent job of explaining rationales given europeans' known tendency of bickering on pretty much everything.
> Facebook's tax of 5k/year in the UK, the double irish with a dutch sandwich, the backroom deals from the 80s with the Irish govt.. these are global practices which truly devastate everyone in the long run.
And are all legal, and designed as so by the government. Auditing won't make a difference, since it is legal. Don't like it? Make it illegal. Simple as that.
> actually trust bureaucrats to step in and audit - and subjectively think they very often do a decent job of explaining rationales given europeans' known tendency of bickering on pretty much everything.
Oh, no issue with that. My issue is with this:
> Or Google and Facebook will be held to a higher standard of compliance.
This would be a valid point or topic of debate should the subject be an actual law (the 70 years copyright law for instance?)
What we're dealing with here is loopholes and nothing more under various incarnations - take patents in the US or the lax tax policies of Malta or Isle of Man as an example - and you'll get a sense on why a bit of cleaning up is sorely needed.
> What we're dealing with here is loopholes and nothing more under various incarnations - take patents in the US or the lax tax policies of Malta or Isle of Man as an example - and you'll get a sense on why a bit of cleaning up is sorely needed.
Suggesting that US companies should be above reproach or oversight, by definition, or that the EU would treat local or Asian companies who tried to pull the same stunts more generously is not based on evidence.
If US corporates try to flout European law, they'll suffer consequences. Market cap does not provide a free pass for being a bad actor.
This is an insightful comment. It costs a lot of money to stay in compliance. One of the charges leveled against the ISO was that it was designed to be hindrance for Japanese companies to compete, because the organization ha(s|d) a lot of influence from Western entities.
I wonder if you know that as of this week the US government can request any data on any user ID an American company even if that user and their data are not on American soil?
(Possibly thanks to GDPR) companies may object to that request if it costs local laws.
But yeah. Go on pretending that the EU lives to target American companies. From a European's point of view, American companies are not find enough as they view privacy, data, sovereignty etc. as some abstract concepts that don't apply to them.
They don't target American companies. American companies makes themselves targets. If anything it is the US that tries to hurt European companies (actually any business that isn't American). If US law can't control a business like Facebook, somebody has to. One can only hope it rubs off on American citizens too so they get a bit more privacy (let's forget NSA for a second).
>> They don't target American companies. American companies makes themselves targets.
Sometimes the only reason they make themselves a target is because they make money, and American companies are the largest players in the space, not EU ones.
Yours is not a unique sentiment, but I find it so disheartening. Vague laws enforced selectively are bad for the rule of law. Is the schadenfreude from sticking it to whatever American company is selected worth that?
No, it's just that American companies as a rule tend to have a dim view of privacy and some pushback against that is welcome.
As for enforcement, if you're a European company you have much more to be worried about since it is going to be much easier to go after you.
Selective enforcement is hopefully going to be limited to going after a couple of very prominent offenders after which the remainder will fall in line.
>Is the schadenfreude from sticking it to whatever American company is selected worth that?
No, but a correction is long overdue to signal to international companies that the primacy of politics still exists. How companies conduct themselves is determined by European law and European citizens, not businesses.
Companies have only themselves to blame for having brought it on. Whether it's skirting taxes, mistreating user-data or assisting in election-meddling, if companies are not willing to self-regulate they will be regulated. That's an overdue message to send.
That is a pretty naïve view of affairs. Practically all the banks even the European ones like HSBC, RBS etc have been complicit in committing all sorts of crime and continue to do so with impunity.
That's too close to what-about-ism for my tastes. I want those banks to be held responsible too, but I'll take just one or the other if I can't get both right now.
It is not the business of companies to self-regulate. Unless there is some kind of certification involved. Eg the gaming industry self regulates by having different ratings for games depending on their content. Such a thing is likely not possible everywhere, and especially for something as general as data, and thus legislative rememedies are required
i've previously researched this idea because it comes up rather frequently.
This was in anti-trust, because privacy-related enforcement is actually left to the EU states and doesn't actually happen very often.
But in anti-trust, there is no evidence of any bias against US companies. The distribution of (the sum of) fines among EU, American, and Asian companies actually tracks closely with their respective market shares in the EU.
The only divergence I found was slightly fewer/lower fines being levied on US companies, and slightly higher numbers for Asian companies. I would attribute this divergence to the far more advanced oversight in the US vs. most Asian countries: the EU doesn't need to get involved if these companies are well-regulated in their home countries.
It'd be interesting to do a similar analysis for the US. As a
German, I've heard similar grumbling regarding the fines against Volkswagen and Deutsche Bank. Personally, I doubt that those fines are motived by nationalistic motives. They might just show different approaches to corporate wrongdoing, with Germany either being too chummy with its industry, or possibly putting too much trust in the force of negative publicity alone to change behaviour.
Privacy will invariably generate more headline-worthy conflicts with US companies, simply because that's where all the large companies are located. But if you're in the US, I'll also point out a certain bias in what you'll see reported, considering EU fines against exclusively EU companies are unlikely to make the US news.
Trust me when I say that privacy happens to be a major concern of many voters in the EU. There is robust and constant debate regarding, for example, data collection by governments and law enforcements. In Germany, there is major opposition against the increased use of video surveillance, the use of phone location data in law enforcement, and EU legislation requiring the retention of communication data.
There is a low-hum undercurrent of anti-Americanism in some discussions of Facebook, Google et al, possibly more so in France than Germany or the UK. But Europeans are constantly exposed to US media and such familiarity breeds trust. If Facebook where Chinese, or Google were Russian, distrust and call for regulation would be far stronger than it is.
> I mostly agree with you, but the penalties for non-compliance are so staggeringly large, the requirements are so vague
It'll be interesting to see how the first few non-compliance cases go. I would like to think if a requirement was vague, a company had made a reasonable interpretation of it and the company was responsive in making changes to become compliant that no fines would be issued.
This is true of certain parts of Google's business, but definitely not with other parts. It seems Google is hoping websites will agree to make Google a "co-controller" of their sites, and get their users' permissions for Google to collect data on them.
The entire ad business, aka, most of their money, is not really GDPR compliant. Collecting tracking data from third party sites is not okay.
> Google has effectively been GDPR compliant or close to it for years
True, sort of. They only did that because the EU started an anti-trust investigation against them, though. And now they risk paying a $2.8 billion fine, so maybe Google is still worse off in the end?
Not really. Google has been following these practices waaaaaay before the EU fine. Both the export tools, logging of all your activities (and ability to actually delete them), and automatic expiration of all data derived from user signals.
I think Apple is so many years behind Google in AI that they have little hope of catching up. Their bet will be that consumers are privacy-focused enough to not mind the fact that their products have less AI-focused advancements (e.g. Google Photos).
At the end of the day, the only way AI systems currently train themselves is if a lot of users share their data, like Google-users do. Apple provides more of a firewall, but handicaps potential AI systems from using training data in doing so.
Also Google has massive mindshare in the AI community with TensorFlow.
I use both Google Photos and Apple Photos and honestly, most of the "suggestions" from Google Photos are uncompelling. When I switched back to iOS a few months ago, I didn't find anything keeping me from moving primarily to Apple's solution. The most impressive automatic-photos-adjustment thing I've seen is in Lightroom now - and Adobe wouldn't be my first guess for AI leader.
I find this to be a general trend with Google: they may have more advanced AI, but haven't yet figured out super-useful ways to apply it.
They also are perfectly willing to undercut their own AI to serve business goals: the Assistant on my Pixel wouldn't even let me do "hey google set a timer for five minutes" if I didn't give Google my full location history. Because you need that to set timers...
OK I guess we have a different experience. As an iPhone user who uses Google Photos I find it to be quite a lot ahead of Apple's solution. It isn't really the filters it's more the search feature. I can search for "pets with children" and it will give me accurate results. Apple doesn't come close.
That would explain it. I do my photo browsing by time, not by search.
I've _tried_ searching in Google Photos before, it just didn't work reliably enough for me try it more frequently (usually searching for more specific stuff than something like "pets with children," though I can't recall any exact examples right now). Searching for location works well, but that's not very hard.
Anecdata, perhaps, but I have noticed that Google's photo searching by content has greatly improved as of late. Searches that failed for me around Christmas of last year now work perfectly.
For me, the killer ML use case in Google Photos is object recognition.
For example, I took a picture of my driver's license a year or so ago, and needed the number on a form recently. I just went to Google Photos and searched for "driver's license." Sure enough, the old picture showed up, even though I never manually tagged it.
The same is done in the photos app on iOS and macOS, all locally on device. I don’t have a picture of a drivers license, but I just tried “license plate” and “receipt” and both brought back all relevant results.
Which is mostly useless when you have only 20GB of storage.
OTOH, if I ask Google Photos to show me pictures of me on the beach with my friends X and Y, it shows me a picture from 2005, of exactly what I asked. (literally "me on the beach with X and Y").
And if I ask for pictures of me and my friend Z in a party, it shows me pictures of me at Z's wedding 4 years ago. (literally "me with Z in a party").
Assume you mean the 20GB limitation being iCloud storage? Both iOS and macOS Photos can “optimize photos storage” and will build index of full library without needing to keep photos on device after they’re analyzed / indexed.
Don't you carry your driver's license in your wallet? It's just weird to me that someone would think "I need my driver's license number, so I'll search through an online photo archive."
I’m not OP, but I’ve had similar experiences. I don’t carry my license in my wallet because I live in a foreign country where my license is useless (we have a different form of government ID), but I do still need it for occasional administrivia back home.
Their advance is because they have two geeks at the helm that have been dreaming to have AI to respond to search queries since pretty much the beginning of the company.
> I think Apple is so many years behind Google in AI that they have little hope of catching up.
We are so early on in the AI revolution, it seems very premature to make this claim.
I would wager we can't even imagine the scope of the kind of data we'd have available in the coming decades, let alone how the latest AI tech will use it, or which company will lead the way with its application.
> I think Apple is so many years behind Google in AI that they have little hope of catching up.
That’s been used to describe every dominant technology company at some point, most of which have been supplanted by now (or else supplanted a company which was previously described in similar terms). Maybe this time it’s true—and Apple certainly has their work cut out for them—but history implies that it’s probably not. Maybe it won’t be Apple, but it’ll probably be somebody.
"I think Microsoft is so many years behind Google in search that they have little hope of catching up"
I can think of more examples.
I agree, and Microsoft is a good example, that a paradigm shift can totally change the playing field. A desktop software oriented company like Microsoft couldn't or wouldn't take their dominance into an internet oriented market.
However Google has managed to transition to mobile and demonstrated they can manage such changes.
From what I've seen of fuchsia they're thinking ahead again.
I don't think this is true. I do specialize in ML and you are probably talking about neural networks ("deep learning"). While they do require a lot of data for most tasks, it's still finite and after a certain size the improvements are not significant anymore (the definition of significant is up to you..). I don't know what Apple is working on, but i think for most of the ML-applications they are interested in, getting the training data is not that hard (like suggestions on the photos-app or autocorrect). It's still apple.
If I remember correctly, sharing analytics info etc. with apple is opt-in? If 1% of the users share the relevant data (or you've got money for annotating data in asia/africa, like tesla), this might enable you to solve most of your problems.
this does not mean that the quality of apples solutions is the same as googles, i just doubt it's the training data. Google is A LOT (not even comparable) more visible in the ML-community and i bet they have better and more researchers.
Not disagreeing with you that Google has advantages in some areas. However, Apple has some advantages as well, especially in its hardware/software integration in end user devices, which allows it to deploy e.g. hardware neural network acceleration or specific sensors, as in the iPhone X.
Is your background in AI/machine learning? Because it is a very common trope that is rolled out and I strongly disagree with it. There isn’t such a thing as an insurmountable lead.
Given the public desire to make AI largely open source - which Apple could easily bankroll and gain goodwill while doing it - I'd say the odds of catching up are fairly good. Not necessarily over 50%, but probably better than it would look from counting patents.
Among all the data-dependent tech companies, I believe Google is one of the best situated to comply with GDPR. Mainly because they have lots of cash to tackle regulations in terms of resources and manpower. Also, Google has revenue streams (Cloud, Play store, non-personalized Search ads) which are not as user-data dependent as some others (eg. FB, startups with ads-based business model).
AI itself is so far behind than most people realize. Most AI-branded products do an extremely small subset of tasks. I argue that it isn't AI rather using arcane statistical methods to hopefully ascertain the human requirement for the context.
AI doesn't yet exist anywhere that I know of. Hiring an AI guy is symbolic move used to signal to investors that they're making moves in a certain space.
>that it isn't AI rather using arcane statistical methods to hopefully ascertain the human requirement for the context
One of the big problems the field of AI has in terms of marketing itself is that is soon as a problem is understood and put in to use, people no longer see it as 'AI'. It's the god of the gaps, so to speak. But that does not render the actual advances not-AI. It is a perceptual problem, not a real impediment.
Google's big problem is straight antitrust, followed by that they're an adtech revenue engine locked in a multi-front war against companies who sell real products. Their GDPR story is actually pretty good and their transition away from adtech revenue seems to be progressing well too.
Do not think Google mhas anything to worry about. They can just tie ads to the content instead of the person. Plus search does not require other data as ads are associated with what you are searching.
Could you elaborate on GDPR being opt-out? The main point I keep reading about seems to be unambiguous separate consent for every use you use any piece of data for. (on mobile, but I’ll come back with references if needed)
Consent is one basis for collection. “Legitimate interest” is another. You need consent when you don’t have a strong enough argument under legitimate interest or another lawful basis.
IANAL either, but my research confirms this also. A lot of companies up to now have used opt-out as their skewed version of consent. GDPR raises that barrier much higher, however, some of those use cases of "older style consent" may very well fall into legitimate interests or performance of a contract.
One of the main things about GDPR though is getting companies to give a monkeys about the rationale for the data collection - even if they use the legitimate interests argument, part of it is that they now have to actively prepare for that, show the legitimate interest and rationalise it to regulators.
Thank you for the keywords, that actually makes some sense.
I guess the crux of the question is in ambiguous terms like "personal freedoms", etc. that the regulation says should be weighted against "Legitimate Interest" of whoever collects data.
He doesn't have an AI background, although I'm sure he's picked up some stuff over the last few years. Google made him head of a bunch of different things because of his project management skills. (I used to work in his department.)
He founded Metaweb but unless I wasn't paying attention he doesn't have a PhD. He's the same age as me so his academic career was in the 80's. jg isn't a big time AI guy but he is very smart and he's been deeply involved with AI projects for at least the last decade.
KG is not what most people think of as AI. It's more of a very large triple database with some confidence metrics than anything like what would pass for machine intelligence today. It's not even in the same org at Google.
He doesn't have any modern AI experience. However, since researchers like Ruslan Salakhutdinov are already working for Apple, Apple already has the tech expertise, and probably wants people who can grow and manage teams, and get the most out of their AI talent, something researchers like Russ are not experts in.
However, I think this move benefits Google as well. Jeff Dean now leads AI, and unlike JG, Jeff has both the technical chops and the managerial expertise to boost Google's AI programs. No more having to approve every project by someone who doesn't understand the nuances and capabilities of modern AI. If Sundar gives creative freedom to Jeff, we might even have singularity by 2050. I am only half joking.
He's founded and managed Google Brain. That doesn't compare with someone who's more focussed on management, but I think only a handful of people in AI have both technical and managerial experience, and Jeff is one of them.
Most people just don't care about the voice assistant on their phones, despite the efforts of Google, Apple and Samsung to make them care. Using a voice assistant in a public place still marks you out as some kind of weirdo. Siri is great for setting timers, that's all that matters really.
I'm honestly not sure what else I'd use a voice assistant for and Siri works great for all of those tasks. Usually when I have an issue it's due to background noise, so hardly Siri's fault.
Quite frankly the "siri is crap" meme seems to be just that: a meme, with very little basis in reality. She's not quite as feature rich as the others but... who cares?
Agreed. Looking forward to when voice assistants can be taught new tricks by regular device users instead of having the new tricks be programmed by developers.
I heard rumors from Apple people that Carlos Guestrin in Seattle and Russ Salakhutdinov in Pittsburgh were both trying to be Apple's head ML/AI person. I wonder how Giannandrea's hiring will effect this.
This is going to be interesting to watch over the next few years, given that Android had a huge setback in the Oracle v Google case [1]. The huge sell of Android over iOS, besides the customization features, was the Google AI superiority over Siri.
My hypothesis is that this move, if executed well, should help narrow the AI perception gap. I think customers know that Google will probably always be superior in AI, but if the gap on mobile is marginal, and combined with the added security that Apple is famous for building into their phones during a time when the Facebook scandal and other data breaches are causing users to be more sensitive to their data, could result in a very interesting market share shift over the next few years.
I'd agree with you, but almost every Youtube video I watch comparing iPhones vs. Pixels, or HomePod vs. Google Home, always tend to get to focusing on what Google queries can accomplish vs. Siri queries. It might not be called "AI" concretely, but the sentiment is probably there to some extent for some users.
Some users, sure. Maybe 10% of all people that look at Youtube videos about new phones will care enough to make it part of their decision. But they're a small minority of everyone that buys phones.
Homepod vs Google Home is a bit different though, since that's their only reason to exist.
The disconnect between median users and the reviewer bubble has been widening for quite a while. But: it's hard to know what median users actually care about, it's not hard at all to know what reviewers care about. Also, median users are not median buyers.
It reminds me of how cars suddenly started to sprout an absurd number of cupholders at the beginning of the century, probably because they are an easy thing for reviewers to write about.
The strength of Google’s AI is not because of its technical superiority (that helps) but because of the huge amount of ubiquitous data it has on users. AI is not magic, it needs data to perform well. Challenge for Apple is where will it get this data to provide customized AI facilities to its users, and this will inevitably lead Apple to the privacy issues Google is facing.
> Challenge for Apple is where will it get this data
iPhone, Safari, Maps, Photos, App Store, iTunes Store, Apple Pay, Apple TV...
They may never have as much data as Google, but they have a lot.
The latest hotness in AI research is self-training, which helps extract more goodness from data. So with some cleverness they may be able mitigate their data disadvantage somewhat. Of course Google works on this area too.
Apple has a 600 GFLOPS NPU in the iPhone X. Nvidia's $3000 Titan V has a 110 TFLOPS TPU. After a couple of decades of researchers using GPU's for compute, dedicated hardware is available, with a better FLOPS/$ ratio. Looks like the future of compute.
We'll see Apple releasing more uses of their NPU for AI.
I wasn't confining it to machine learning, but compute in general. (I'm interested in simulation.)
Are machine learning benchmarks applicable to non-machine learning computation?
I'm not understanding your comment correctly, is it sarcastic? Nvidia doesn't ship a TPU on the Titan V. Also, 0.6 TFLOPS/$1000 vs 36.7 TFLOPS/$1000 is nearly a 2 orders of magnitude difference favoring Nvidia.
It's not sarcastic. It's not contrasting the two, but pointing out a general shift towards NPU/TPU (as the comment states). Of course a mobile device isn't going to have the same power ratio as a Titan V (It's amazing that the iphone X has 500GF, in addition to its GPU and CPU. I expect it will go much higher; this is just an initial foray).
I don't understand the objection to describing the tensor cores of the Titan V as a TPU. As a GPU, it has only around 10TF. How would you describe the tensor cores?
Is Giannandrea an "AI guy"? His background, albeit impressive, suggests he's more of a search guy who got AI added to his plate of responsibilities. Maybe that matters, maybe it doesn't.
Anyone know? I'm just wondering if Apple hired a fine executive, just maybe not the best person to run AI.
And yet, Intel actually had a written document describing a non-poach agreement with Pixar, where Intel's policy was to not hire any Pixar employee without the Pixar CEO's express approval.
"If a Pixar employee applies to Intel without being recruited by Intel, contact Pat Gelsinger and explain to him a Pixar employee (provide the candidates name) has applied to Intel without being recruited and he will contact the CEO of Pixar for approval to hire," read the Intel document.
...
"Mr. Jobs wrote: "I would be very pleased if your recruiting department would stop doing this." Mr. Schmidt forwarded Mr. Jobs's email to undisclosed recipients, writing: "I believe we have a policy of no recruiting from Apple and this is a direct inbound request. Can you get this stopped and let me know why this is happening? I will need to send a response back to Apple quickly so please let me know as soon as you can."
Mr. Geshuri told Mr. Schmidt that the employee "who contacted this Apple employee should not have and will be terminated within the hour." Mr. Geshuri further wrote: "Please extend my apologies as appropriate to Steve Jobs. This was an isolated incident and we will be very careful to make sure this does not happen again."
Three days later, Shona Brown, Google's Senior Vice President for Business Operations, replied to Mr. Geshuri, writing: "Appropriate response, thank you. Please make a public example of this termination with the group."
This guy is an exec and like all execs he is good for shallow view from 35,000 ft. He isn’t in the field and no one I know who is in the field seem to know even his name. I have zero clue on what is his contribution to the field let alone what kind of actual technical details he knows. As far as I have known these sort of people, they wear suits, memorize buzz words that their reports talk about and take credit of work done by their reports. Such “AI exec” hire is no-op in my book and it isn’t a surprise that Apple has to run PR articles to justify their purchase of a suit.
Some people on HN seem to constantly despise any manager in a company that is not a technical person. Whether it's a majority or not is hard to tell, but it's a conspicuous group of people for sure looking at upvoted comments of the sort.
The theory seems to be that all these companies work only because of their engineers. Management is just useless. They are just "people with suits, that memorize buzz words take credit for the work of others".
With all such incompetent people in high position, one has to wonder how these companies seem to survive for so long.
Maybe, just maybe, they are not so incompetent and raise to those positions because of merits. Maybe engineering skils are not the only needed ones.
It seems to me that this culture is just based on resentment for the people that are higher in a hierarchy. From this point of view, the only way to feel good is to disparage them and assume they are there only because of the "corruption of the system".
No, my friend. I have lived under and interacted with enough suits at big 5 that I can tell you that if 80% of those people disappear you will see absolutely no change in the bits shipped by that group. How do I know this? Well, I have seen one suit getting replaced by another random suit all the time. There have been suits who never actually tried out the product they supposedly owned and preached to outside world. I am not kidding you. They literally had no account in the system. Most suits had no clue where source code existed for their product and even never had read documentation that their customers had to suffer through.
Their known contribution included setting up exec retreats were people are supposed to tell them about what needs to be done and then they go away with “blessing” the plan that got surfaced after their reports fought it out. Another of their contribution was to do meetings with other suits at customers/partners. Another was to do budgets and when things go red (or pressured to show profits), do hard decisions, aka, throwing 10% of team under the bus in form of layoffs. Their next contribution was to present works of their teams when it got done and credit it to their leadership and vision. I am talking about CVP+ levels, BTW.
It’s easy to identify these suits. Go to the leaf level employees in their group and ask them about actual contributions of their CVPs in their product that made a difference. If they cannot mention anything concrete other than blanket buzz words like “leadership”, “vision” etc (in case they cannot honestly answer) then you know you have a suit at the top.
These people are very different from Jobs, Musk, Zuke, Jenson Huang etc who are absolutely down and dirty at every level, knew more details then any individual employee and had real technical contributions in their product that made significant difference.
You are clearly an example of what I was talking about.
You judge the utility (or lack thereof) of these positions while never having had one yourself. Luckily, it's the top management of a company that choses who these are and not engineers.
It's clear from claims like "if 80% of those people disappeared you would see absolutely no change" that you have no idea what their role is.
When you say "most suits had no clue where source code existed" you just fail to realize that that's good. It's not their job to know this, it's an engineer's job.
You talk about their "known contribution" as if it's a general sentiment, but show no proof of it being other than your personal opinion.
In fact, it's clear from statements like " Another of their contribution was to do meetings with other suits at customers/partners" that you have no idea about how valuable talking to customers is for a company.
And tell me, in case you had to "make hard decisions" like "throwing 10% of team under the bus in form of layoffs" what would you do instead? That's exactly one of the cases where the skills of an engineer are the last a company needs.
Bad leaders can ruin great teams and great leaders can turn around weak teams. I've seen/done it myself. Without fail, when I switch out a weak manager for a better one, the team improves. When groups have no manager they might produce, but usually not well or as well as they could.
Management isn't around because some conspiracy made by managers to perpetuate themselves, but because it works. No effective group of people is truly leaderless.
Pure managers (ie non technical managers) tend to hire other pure managers. They tend to over-amplify role of management vs technical problem solving. They often believe later is just “commodity” while projecting management skills as rare mysterious abilities. It is not an accident that pure managers at CVP+ levels are able to command more compensation than entire team of 20 people. This phenomenon is purely because pure managers keep projecting management as some sacred priesthood. In all teams I have worked I have known people who can easily do CVP job at quarter of the cost and twice the efficiency and technical chops. Despite this you see these inefficient people taking away massive funds. Why do you think there is no conspiracy?
That's bad managers hiring other bad managers. It's not a conspiracy; it's incompetence.
The problem likely goes all the way to the board of directors who should hold the CEO accountable to build a culture that doesn't let bad managers exist in the org (by coaching or removing them).
A lot of companies have this problem, and it's a major dysfunction, but make no mistake, it's :bad: managers you have a problem with, not "pure" ones. I know a ton of great non-tech leaders who do amazing work of leading technical teams because they build the team around that. You get the lead engineer(s) to run the tech leadership while empowering them with autonomy / resources / support and holding them constantly accountable to well defined goals.
Management isn't rocket science, it's just a lot of hard work and discipline. It is its own craft and a very challenging and rewarding one if it suits you.
> And tell me, in case you had to "make hard decisions" like "throwing 10% of team under the bus in form of layoffs" what would you do instead? That's exactly one of the cases where the skills of an engineer are the last a company needs.
Layoffs are an effect of organizational dysfunction. In which case, the responsibility lies squarely on the shoulders of management. It’s just that there is a perverse incentive structure in most corporations, so 10% of the team gets laid off to continue to pay for the exorbitant bonuses and lifestyles of management.
You are exactly the type I am talking about. For some reason these management types think of themselves as worth 10 engineers bestowed with some mystical powers called “managing” while having little clue about what was supposed to be produced. This is typical MBA culture and interestingly you also fail to enlighten all of us low links about what your role was supposed to be. Also, noticed that I talked about meeting other suits at customers/partners. This is not same as actual customers. Suits typically only meet other suits. Have you seen how badly designed and almost unusable some websites like banks and Fortune 500 are? That’s typical example of suit never using his/her own product, let alone meeting actual customers.
BTW, knowing where your code is and how to build it is important. It tells you basic health about your engineering process. A good technical manager instinctively knows that anyone should be able to build a product or there is something very wrong. A suit will delegate and hope everything is workings out. If you managed sausage factory and if you had no clue where your factory was, that would be odd. Right?
About the “hard decisions”, suits needs to hold them accountable for those results. This could be done either by getting mega salary cuts or even just resigning for producing a massive failure. I have rarely seen suits getting affected when they had to make “hard decisions”.
> These people are very different from Jobs, Musk, Zuke, Jenson Huang etc who are absolutely down and dirty at every level, knew more details then any individual employee and had real technical contributions in their product that made significant difference.
Jobs isn't a great example to make because he wasn't an engineer either and he very much "took the credit for the work of others" (as you put it earlier).
However Jobs is a great example of the value of "suits" when you have an exec who is actually bloody good at what they do. I've worked with project managers and CEOs who have been excellent at their job. I've also worked with some who haven't. And some of those who didn't work out was more down wider company politics than their own ability to perform.
You see, you cannot make accusations that all execs are the same pedigree of worthless cannon fodder. In fact many of them are developers and engineers, like ourselves, who have risen through the ranks (either via successful startups like Musk or through the more traditional "hard grafting"). Some are project managers et al who have proven themselves. And yes, some are just useless - but you get good eggs and bad eggs in all industries.
Jobs was a GREAT engineer. He was hired to do electrical circuit work for Nintendo and despite the fact he smelled horribly (because he rarely showered) and wore flip flops, he was still called on by customers to solve their problems reliably. All great managers are deep down in details and know their products inside and out, probably better than any individual engineer. My definition of suits are managers who feels knowing details of their products should be left to engineers and their job is, you know, “managing”. Scully was a classic suit. Jobs wasn’t.
Jobs was hired for a Breakout game for Atari and got Woz to do most of the technical work on that job for him.
"Steve didn't ever code," writes Wozniak. "He wasn't an engineer and he didn't do any original design, but he was technical enough to alter and change and add to other designs."
> if 80% of those people disappear you will see absolutely no change in the bits shipped by that group
For a short to medium time, sure (and I think the better the exec is the longer his team will be able to work successfully without him). In my experience most successful execs build teams that build good products.
Remove the exec and the good team still builds good products, but it usually starts rotting (e.g., good technical people are frequently not good at handling people issues), then snaps as good people leave and what is left cannot build anything anymore.
Most good exec contributions are not visible at the technical level (there are some exceptions as you cite), but they still create an environment where smart engineers find it fun to work productively. It feels like some magic is involved in this -- I do not have this skill, but I have seen it in action a few times. Just my 2c.
> I can tell you that if 80% of those people disappear you will see absolutely no change in the bits shipped by that group. How do I know this? Well, I have seen one suit getting replaced by another random suit all the time
This is a feature of a well run business, you want to be able to change your executive teams without negatively impacting on the rest of the business in the short term.
> These people are very different from Jobs, Musk, Zuke, Jenson Huang etc who are absolutely down and dirty at every level, knew more details then any individual employee and had real technical contributions in their product that made significant difference.
Sure in the early days of the product this is the case, but these are exceptional individuals who have transitioned into executive roles over time. They have spent the time since building up a management culture and organization capable of running a billion dollar company, which is not easy to do, and is more than a full time job.
I think you grossly misunderstand what executives do, both long term and day-to-day. At the end of the day a company is mostly just a group of people, with a set of assets. To grow and evolve beyond the current products you need to cultivate this group into a team that can consistently bring success.
Ok, so if team can work without suit then it was because suit was doing great job. So you are saying we need to pay suit 20 times an engineers salary so he doesn’t have to be there and team keeps producing. I get it now.
You're providing the perfect type specimen for what OP was describing...
To try another avenue to convince you: Presumably, Jobs(?), Musk, Page, and Brin were, once upon a time, technologists just like you. It's somewhat certain that at least some of them shared the mindset you're showing.
Considering they ended up hiring lots of managers, it stands to reason they changed their opinion at some point.
I tend to value the opinion of people who have shown the ability to change their mind over those that never have. Add the credentials these people bring, and I'm tempted to accept the idea that managers have something to add to the success of software companies.
I am not against having managers and hierarchy. I am against the idea that one can hire a random suit and that suit doesn’t have to know much about either the sausage or how the sausage is being produced. And all the credit of producing sausage will still be attributed to the suits for their “leadership” and “vision” along with salary worth 20 people who actually put in the work.
Well, most software engineers don't despise managers. Engineers despise suits who don't understand software engineering. Sadly, most suits fall into this category and to make matters worse they turn out to be control freaks. They want to control something that they don't understand and this pisses engineers off. Managing a team of talented software engineers is more art than science. Good luck trying to win the hearts of engineers while being a control freak who doesn't understand what the team is doing.
All the prominent CEOs and execs who stand out are exceptions to this case. They foster strong team building and give complete freedom for the team to execute. They own up to their mistakes. Most of them are not control freaks, but they are detail oriented (Jobs). It's a joy to work with such managers.
That's funny to see the whole range of opinion about management on HN. It varies from this one, the very bottom, the leech living of his team of engineers, to Elon Musk, the engineering genius whose team(s) is acknowledged at best as an afterthought.
I'm not in "the field" but this article certainly makes him sound like he is involved with machine learning / smart assistants / etc. Is this really a "shallow view from 35,000ft"?
This is classic PR release. If you are a suit (CVP+ levels) at big company, you are entitled to 2-4 PR releases like this. Lowly VPs typically get only 1 PR release per year. Typically the job of PR release is to portray as leader and for that purpose all work done by their underlings including past successes gets attributed to them. Journalists have neck for amplifying such portrayals. Now that you know this, hopefully you can point out such hit jobs in future.
Ok but you haven't answered my question - does he really have a "shallow view from 35,000ft"?
You said "He isn’t in the field and no one I know who is in the field seem to know even his name" but the article I linked to seems to clearly indicate he is in the field - can you reconcile the two?
You also said "I have zero clue on what is his contribution to the field" but the linked article does seem to cover this. Is it wrong?
What contribution are you talking about that is in the article? In the article, he talks about StarTrek and HAL5000. Anyone can do that talk, especially the posers. It’s typical zero-information PR article written by suit, written for suits. People in the field can intelligently talk about details of recent advances, exactly what techniques enabled them, what are the disadvantages, what can be realistically done etc. People in the field typically have written a paper or been part of at least failed experiments with technical contributions. He happened to be suit managing smart speaker project. That’s an administrative job and doesn’t count as being in the field, let alone “AI chief”.
To begin with, it shouldn’t be hard to see that this article is written by Apple PR and then handed over to journalists. This guy was never Google’s AI chief, that person - if there is one - is Jeff Dean. It’s strange to see Apple parading around a suit who basically managed hodge podge of projects at Google and calling him Google’s AI chief. Go ask around anyone who has anything to do with AI and drop his name and you will get blank stares.
However I would argue hiring a 'suit' who was CTO for a reputable knowledge based AI company 8 years ago, and since been in a strategic position during Googles recent progress is more valuable to Apple at this time than a well published researcher.
Define “good management” as relavent to products requiring creative work. Most frequent definitions I have came across included “good listeners” and “delegators”. Much of these can also be accomplished by a statue of a monkey god.
My impression is that he's very technical, but his background seems to be in search, not AI.
Perhaps there's AI work in his background that isn't evident just looking at his LinkedIn page.
Perhaps he doesn't have a long-term AI background but has come up to speed since getting AI put on his plate at Google. Or, at least come up to speed sufficiently to be able to distinguish between bullshit and non-bullshit.
Given this story [1], it will be interesting to see how the differences in culture and openness will allow to perform well inside the company, especially if it requires deprioritizing a walled-garden ecosystem to succeed in AI.
I'm curious how this will impact Siri's development. My understanding is that Siri has been (maybe still is?) somewhat ahead in terms of understanding language flexibly and naturally, so if Giannandrea can help them catch up on the things that Google is good at, could they leap ahead?
My uninformed impression is that Google Assistant is exceptionally good at the things Google-the-search-engine is exceptionally good at: Type/say an awkwardly worded query and get a highlighted (and usually correct) answer.
It's not clear to me how Siri gets as good at that kind of thing without the Google search index (and all the results quality work that has been Google's bread and butter for over a decade).
Was I the only one who read the headline as "Apple Hires Google's AI Chef"?
The Google cafeteria has gotten seriously high-tech: now they have an AI-powered chef cooking meals. It knows what you want to eat before you do! The AI Chef is so popular that Google is now selling AI Chef services to other corporate cafeterias.
Machine learning comes up atleast once-a-keynote, so it's not surprising to see a high-profile hire in the area.