Jeff Dean takes over as Google’s AI chief | Hacker News Reader

Jeff Dean takes over as Google’s AI chief

theverge.com

290 pointsquazar133 comments

Seeing the comments, I can't but imagine what kind of personality is needed to not be affected by that kind of reputation and recognition. Like everyone expects everything you do or say to be really smart. I guess its kind of like being a celebrity; but instead of looking or acting good you have to be smart all the time.

Edit: Also, how is Peter Norvig not Google's AI chief???

Norvig's title is "Director of Research", presumably this is a different position than "AI Chief". Source: http://www.norvig.com/

And to be clear, he is a Director of Research, not the Director of Research. There is more than one.

Why would he be?

> I can't but imagine what kind of personality is needed to not be affected by that kind of reputation and recognition.

The reason many Harvard grads keep the fact that they went to Harvard to themselves is because they have a similar expectation -- that everything they say will be brilliant.

https://www.youtube.com/watch?v=q51ld-scMI8

Ouch :-)

"Edit: Also, how is Peter Norvig not Google's AI chief???"

I would imagine Norvig as more a thinker and less of a doer. I think Google has reached a point where their flavor of AI works sufficiently well where now the only problem left is getting buy-in from as many teams as possible within Google.

Jeff Dean might be more decision-maker, politically connected(internally at Google) and less of a thinker-type in moving this initiative forward. They're both extremely valuable just in different ways.

It's one thing to build it(AI) but an entire other problem to get the AI into Google's huge suite of internal/external systems and products.

> Also, how is Peter Norvig not Google's AI chief

He is not doing any pratical AI stuff at this moment. Jeff is still productive AF even approaching his 50s.

That's not a great thing to say. Why wouldn't a 50 year old be productive?

I'm inferring they meant productive in practical tech role, as opposed to moving on to management or leadership roles, which most people tend to do. It's admirable, managing burnout.

Nothing to do with ability.

Burnout, management, contentment with prior success, etc.

I'd love to see a long coding stream from Jeff or another senior Googler, that demoed all the tools they use and the infrastructure they have available to develop and test new code...

Not exactly a coding stream, but a good overview of the tools he uses regularly https://usesthis.com/interviews/jeff.dean/

FYI Jeff Dean gave a lecture at YC on AI last summer:

https://www.youtube.com/watch?v=HcStlHGpjN8

I read this and thought - that's going to be a big culture shift, Jeffrey Dean Morgan taking over Google's AI: https://pmcvariety.files.wordpress.com/2016/04/walking-dead-...

I think an affinity for "brains" still holds here. ;) [as long as my zombie allusion holds...]

Jeff Dean is arguably one of the top 3 most productive and influential industry programmers of all time.

He certainly has the reputation. Are there any stories of people who've worked with him in person floating around (sans Chuck Norris jokes)? Would be curious to read. Have seen stories of Carmack and guys like that have cemented the legend.

A former co-worker of mine was Jeff Dean's "mentor" at (I think) mysimon.com, where Jeff worked before google. He said that when Jeff started, he showed him how to check out code, and then basically just listened to Jeff typing furiously for about 6 weeks. Then, he gave a presentation to the eng team that was basically like "Ok I sped up the system by 20x, here's how." Then he quit and went to google.

Jeff Dean was at MySimon for 6 months: https://www.crunchbase.com/person/jeff-dean-2#section-jobs

Looks as though he got curious about the startup world after DEC, and MySimon had nothing to suit his ambitions.

Ah, ok, 6 months then.

Just curious, who do you think are the other two?

I'd definitely put John Carmack on the list, probably as #1. And I think antirez should be somewhere up there as well

While antirez is a great programmer, I think his accomplishments are far from that of Dean or Carmack. There are many notable folks in his playing field.

I disagree. For example, homebrew isn't the best package manager. But its vast reach is a feat of its own.

Same with redis. Is it the best DB ever? Probably not. But it is definitely used quite a bit in the industry and a lot of people get acquainted with nosql databases and k/v stores through redis.

Engineering accomplishments without proper marketing, education and adoption are quite useless, for better or for worse

Sanjay Ghemawat is another one.

Ghemawat should definitely be one. I sometimes wonder why Dean is teated on a wholly different level than Ghemawat. If you look at their works, almost all of them involve both. Probably someone with more knowledge can shed some light here.

I think I've heard "Jeff and Sanjay" as often as or more often than I've heard them regarded separately. Who invented protobuf? Jeff and Sanjay. Etc.

In my experience, hearing people treat "JeffAndSanjay" as a single entity is way more common within Google than it is industry-wide.

I'm not in Google, I hear of Jeff more often. I occasionally forget that Sanjay exists until reminded.

> I sometimes wonder why Dean is teated on a wholly different level than Ghemawat

Racism.

In part, at least subconsciously and since Jeff Dean is a more soundbyte-friendly name than Sanjay Ghemawat (Although "Ghemawat" is an awesome term for a unit of brainpower, akin to the "Wolfram" unit of ego.)

Also, Ghemawat is less of a public speaker and is not a manager

Sanjay makes fewer public appearances.

Jeff Dean memes

Inside Google, they're roughly the same. You used to refer to them together, they sat next to each other, worked on the same projects as you point out, roughly until Jeff started Brain. Outside of Google, Jeff is more popular probably because he gives all the talks and interviews (plus his legends...). I'm sure Sanjay has presented a few times, but I can't recall any, while Jeff has been everywhere and his slides are linked to all the time, e.g. the latency numbers all engineers should know.

Heh .. I think I was in the audience at LADIS the first time he presented that slide (or maybe the first time I had heard of it). Feel really privileged.

Not the OP,maybe Bill Joy and John Carmack.

Another nominee is Doug Cutting

Anders Hejlsberg?

Yes, sure. Turbo Pascal, Delphi, C#, .NET (I think I read that he worked on it too), TypeScript. Maybe LINQ too.

Bill Atkinson, maybe?

Fabrice Bellard - https://bellard.org/

In a leadership role, it will be difficult for him to contribute technically IMO. I wish he remained a programmer, and continued using his abilities to solve greater things.

I am sure all the people he will be leading are passionate, he may not have to deal with the run of the mill org issues. Being promoted gives him more influence/power to implement his vision.

BTW, he is Jeff Dean, he can be the AI chief and hands on technical at the same time :) May be they promoted him because he was bored at work.

Gives him something to do while his code complies.

They promoted him because they already have AGI and it was bored at work.

Jeff Dean is Google's AGI

Na, it makes perfect sense, especially if the person isn't socially retarded or unfit for management. They're able to contribute meaningfully to broad architectural designs to have a larger impact then hacking away at a single project, which also imparts their thought process to a larger group of developers who may end up in his shoes one day. Ideally someone in his shoes can recognize top talent and ideas as well, and clear organization barriers away for individuals and groups to achieve large things as well. Besides, people with his ability and experience should know everything that is going on to influence what the company chooses to do with its money and time over a political bonehead.

He was already in a leadership role (he led Google Brain). Now he's just in a bigger leadership role.

This is a myth spread by people who can not code.

Have you read Coders at Work [1]? There are 15 computer scientists interviewed (listed below). Perhaps it's time for another edition with Dean, Carmack, and others.

Coders at Work Interviewee List:

- Jamie Zawinski

- Brad Fitzpatrick

- Douglas Crockford

- Brendan Eich

- Joshua Bloch

- Joe Armstrong

- Simon Peyton Jones

- Peter Norvig

- Guy Steele

- Dan Ingalls

- L Peter Deutsch

- Ken Thompson

- Fran Allen

- Bernie Cosell

- Donald Knuth

[1] http://codersatwork.com/

interesting to note that Jamie Zawinski, Peter Norvig, Guy Steele and L Peter Deutsch are all lispers

So is Peter Seibel, the interviewer…

And Zawinski [1] achieved what he did with only a high school "degree" (in terms of formal education), according to what Norvig said here:

http://norvig.com/21-days.html

https://en.wikipedia.org/wiki/Jamie_Zawinski

4 of these guys are Google employees right now:

- Brad Fitzpatrick - Joshua Bloch - Peter Norvig - Ken Thompson

Josh Bloch left Google several years ago.

I had seen the trend, via tech news, from some years ago, of Google hiring many top CS people. They might even have the largest number of such people of any company in the world, maybe except for a few other possible toppers like IBM or Microsoft? Just guessing here, based on some anecdata.

And while I'm guessing, let me add a bit of imagining :) I used to, and still wish, that Google or a company like it would come out with a RAD GUI app creation tool like Delphi (of course better is fine too) and make it cheap or free to use to develop GUI apps with (and then keep supporting and developing it, and not make a mess of it like Borland/Inprise/CodeGear did - at least their management did or seems to have done, based on news read over the years). They definitely have the brainpower / bandwidth for doing such a product. But instead they are web-and-mobile-only (or mostly, AFAIK), unfortunately.

>like Borland/Inprise/CodeGear did

By somewhat frequent changes of direction, I mean, at least some of them, particularly during the Inprise years. Like changing focus from dev tools vendor to ALM vendor.

Brad is amazing. He has contributed so much back to the OSS and written excellent code. A true hero, also the youngest of the lot in the book.

I've pretty immense respect for Brad. He's very patient and empathetic in how he interacts with people on email lists, etc.

I would consider myself a success if I have even half of this empathy and patience. :)

Just a humorous thought: imagine getting all of their resumes for a software engineer position. How do you even interview these people or even choose one for that matter?

You stand, salute and offer your chair, obviously

"Welcome, Dr. Knuth! Let's see if you can invert a binary tree..."

Judging by how ridiculous hiring (still) is in the software development world these days I suspect a lot of companies would give them the full fizz-buzz treatment if they made it through the first 3 phone screens.

If you like CAW another good one is Masterminds of Programming.

The author of the "Coders at Work" was asked why he did not have interview at Carmack in one of the talks and his reply was he had contacted him but Carmack did not reply back and he said that with a very bad attitude as though Carmack owes him.

Would Dave Cutler make your list ;)?

https://news.microsoft.com/features/the-engineers-engineer-c...

Right. He basically made AMD64 happen because he didn’t like Itanium.

And has a major OS to his name.

If you mean NT, then he also worked on VMS before it, right. So I remember reading in the book Inside Windows NT.

Yes, he should be in any top 5/10 list of most influential software engineers.

Such a shame then that that talent is focused on making people click ads

Like Bigtable, MapReduce and Tensorflow haven't had any impact outside advertising...

That's true, but if not for advertising they wouldn't exist.

Advertising is like pornography: lots of technology finds its first application or even invention in either.

They open source all their excellent work. Rent seeking cable providers are trying to take advantage of the net neutrality repeal to make people click ads.

It could be worse, a couple generations ago the best minds were used to find new ways to kill people.

That seems like an uncharitable reading of history. Even if technically correct, this interpretation of the effects of nuclear weaponry on the number of humans killed in war seems wildly off-base.

https://ourworldindata.org/war-and-peace

In the short-term anyway (by which I mean 1945 to the present day), the advent of nuclear weaponry seems to have decreased the number of humans killed. And that was certainly the most reasonable rationale for the use of nuclear warheads to end WW2.

You're discounting the non-zero chance that nuclear weapons will end the world as we know it. You can't say that about clicking ads.

Still, it's ultimately about balancing alternatives.

Whether there is a net-gain in exchange for an extremely rare potential scenario which every party is massively disincentived from ever pursuing even a minor exchange of 'tactical' nukes... let alone the even smaller probability 'world ending' volleys. War was brutal for centuries and was only getting more destructive until nukes.

As much as Oppenheimer may have regretted his involvement it's still has not turned into the total disaster many predicted. We could certainly do more to minimize the threat by reducing stockpiles but from a pure philosophical view of having nukes or not...eh I see a good argument in favour of it.

Likewise people clicking on ads has generated tons of value for society and is a big foundation of the recent technical progress we've all benefited from. This is the tradeoff we've exchange for the relatively minor downside risk of ads making people engage in meaningless consumerist pursuits or unhealthy behaviour or be lied to slightly more efficiently by politicians.

Your analysis has a number of assumptions. First that anyone contemplating nuclear weapon use is considering net-gains. Second that every party is disincentivized, either in rational calculative terms or in terms of their perception of the incentives. As a counter I give you the modern USA - what is the disincentive for the USA to use 20 or 30 tactical nukes if that achieves one of its strategic aims? Thirdly, what if I told you that the probability of a world ending volley approaches 1.0 ? What is the evidence that it doesn't?

At the height of Watergate a drunken Nixon was quoted as saying "You know what the difference between us is? I'll tell you, I can walk out of this room right now and give an order and a hour later 80 million people are dead". Cuba was nearer the real deal that either side realised, we've had launch orders ignored and lose weapons that were a fail from going off - which almost certainly would have been interpreted as a strike.

And yet people construct everything as ok.

> At the height of Watergate a drunken Nixon was quoted as saying "You know what the difference between us is? I'll tell you, I can walk out of this room right now and give an order and a hour later 80 million people are dead".

Sorry, I'm really struggling to take this example seriously, despite the fact it was obviously meant in jest and Nixon was far smarter than most people give him credit for, as it ignores so much context, obvious deterrents, and basic common sense that it hardly changes the equation at all.

Nor does the two potential major known nuclear near-accidents, as I've already conceded that much could be done to minimize the risk. Yet it could still be entirely logical to maintain nuclear weapons as an option going forward and to hold the opinion it was historically ultimately beneficial to society as a whole...

Just today on HN there’s a story about another ad-driven company and their role in genocide...

AI and genetic engineering also might end the world as we know it.

I know nothing about genetic engineering, but I know a lot about AI as it now exists. In an imaginary future it is possible that AI could become a technology that adds significant danger, but now the most dangerous element of AI that exists is definitely the bayesian reasoners that are used in nuclear fuzes, and without the nuclear bit they would not be all that dangerous would they? In terms of AI as a danger you can substitute "12 year old boy" in the loop and get a nastier result every time.

Sadly and wickedly many horrible people all over the world have reached the same conclusion.

Not that much different today, a significant portion of high quality talent is still being drawn into the defense industry. But largely to destroy people's privacy and exploiting the weak state of information security instead of killing people.

Aren't you forgetting the vast majority of Google's products that aren't ads, like Search, Maps, Android, Gmail, etc.?

Those make Google money almost exclusively through directing eyeballs towards ads.

There is a difference between product and business model.

I understand what you are trying to say. But all the products you mentioned generates revenue through ads, both directly and indirectly. Google still makes more than 90% of their revenue from ads.

Yes, they are trying to diversify for some time now, maybe they will succeed. but they are not there yet. Most of their current brainshare works on products that make money from ads both directly and indirectly.

Is it wrong if a hospital makes money from ads?

I would hope so, considering they probably would be sharing patient personal information for relevant ads.

We're not talking about where revenue comes from, we're talking about what products engineers are working on. The vast majority of engineering effort at Google is on products like the ones I mentioned, not ads, and it's insulting to Jeff Dean and others to imply that they're not doing anything other than serving up ads. Search, Maps, Android, etc., are all incredibly useful and technically challenging products, and Jeff Dean's code and inventions are used in all of them.

I understand what you are saying.

But the issue is not so black or white. Google/Alphabet is not pouring billions of dollars on free products from the goodness of their heart, they are doing so increase their revenue from ADs. As matter of fact, I would think that most of their ad revenue does not come from direct ad products, but indirectly from other products.

Whether ads are ok or not is a different argument (I don't have any problems with Ads as long as they are not too annoying), but the feeling of righteousness is something I can't understand.

Please understand Google is a for-profit public-traded company, with the only objective of making money for their shareholders (nothing wrong with that), but to imply they have all these free products without any hidden revenue agenda (is, mostly Ads), is just ridiculous.

I don't think @CydeWeis was dismissing the revenue motivation, but rather stating that the engineering resources themselves are being poured into other innovative products that users love, not just into AdTech, even if these products are primarily motivated by ad revenue for Google, the company. Individual engineers, like Jeff Dean, may also have different motivations than the publicly traded company for doing the work they do. It's likely the engineers working on Maps, etc. and the underlying tech would not see their work as simply "AdTech."

It's absolutely very likely that individual, well-respected developers, and engineers are working on cool things in the company that directly has nothing to do with ads. I have nothing against those individuals or those products or even google for that matter.

But the business model is much more complicated than that.

Their product is their vessel for ad money, you can't mention one without talking about the other.

So when someone says: > vast majority of Google's products that aren't ads, like Search, Maps, Android, Gmail.

It's a very simplistic view of how Google business model works. Because AFAIK, only 3 Google products are directly ads related, but ~90% of their other products exist for the sole purpose of driving business to their ads products.

ie, Machine learning in adsense: https://adsense.googleblog.com/2018/02/introducing-adsense-a...

> but ~90% of their other products exist for the sole purpose of driving business to their ads products.

This is flatly false. While certainly, many of the products are used in ads, yes, very few products exist for the sole purpose of driving ads business. A few examples of things that don't: Drive, Docs, Gmail (GSuite/B2B), GCE (B2B again), ML, TF, etc. (again, B2B offerings, as well as being used for all sorts of internal things that have nothing to do with ads), Youtube (Red, TV), etc.

While many things certainly drive ads revenue, there are very few things that you could argue exist solely, or even mostly, to drive ads revenue.

(Am Googler)

But I never said 100% of Google products exist to drive revenue to Ads, I said most do, ie, most free products.

We are not talking about B2B or B2C products for obvious reasons.

Outside of B2B and B2C offering (ie, Gsuite and Youtube RED), both youtube and Gmail makes money from ads. Docs is part of the Drive and Drive with Gmail share same free space limit. Drive is free tier to their B2C offering (pay money for more space). Just like Dropbox has a free tier. Docs is a feature of a drive - which they needed to compete with other similar services.

TF is used as a paid service (along with TPU) for their GCE, also TF is a great marketing tool to hire top-notch machine learning enthusiast to their company. Nothing wrong with that.

According to this (not sure how accurate this is): https://www.statista.com/statistics/266471/distribution-of-g...

Google still accounts close to ~90 of their revenue from ads. The good news is that it looks like their diversification plan is working.

I have nothing against Google or their employees, I think it's an awesome company, I would rather live in a world where Google exists. Regardless of their profit motive, it would be hard to ignore their contribution to make the internet a better place.

But "100% of Google products exist to drive revenue to ads" would probably have been more correct than what you originally said, which was

>but ~90% of their other products exist for the sole purpose of driving business to their ads products.

Machine learning at Google, for example, is a product that exists to, among other things, drive revenue to ads. It is not however a product that exists solely to drive revenue to ads.

As soon as you say "we are not talking about B2B or B2C products", you've made a tautological argument. What you're saying is that "all of the Google products which don't make money via other methods make money via ads". Which is trivially true, but also totally uninteresting.

My point was that the original statement you made was a hyperbole: there are not very many Google products which exist, as you stated, "for the sole purpose of driving business to [ads]". Most Google products have other revenue generation methods. That's all.

One thing which I have adored about Google is X labs ( now part of Alphabet.) very few people are like Larry,Sergey who have the audacity to diversify. They are highly scientific-minded, I see Google/Alphabet both as a tech co and a research , experimentation co. That's what differentiates them from others. Sure, when they fail their success% goes down, but the knowledge acquired by scientific community goes up as a lot of Googlers go on to start their own cos.

I'm not talking about the business model. I'm talking about the products themselves. We're talking past each other if you can't understand that. The majority of the engineering effort at Google goes into features on the products themselves, unrelated to ads. We're talking about the contributions of individual Google engineers, not the business model of the company as a whole. This isn't "overly simplistic".

It's likely the engineers working on Maps, etc. and the underlying tech would not see their work as simply "AdTech."

What’s that Upton Sinclair quote?

Of all the hate that Advertising gets, how will any business survive without Advertising? How does an ad free world look like? You pay for every service as a ad-free subscription model? But how would you come to know such a service exists without Advertising?

But how would you come to know such a service exists without Advertising?

If you need a product you will go and look for it. Advertising by definition is to trick you into buying things you don’t need.

>Such a shame then that that talent is focused on making people click ads

Alphabet's Project Loon was providing connectivity to Puerto Rico in the aftermath of Hurricane Maria. What were the other SV companies, that did nothing to help out, working on?

https://arstechnica.com/tech-policy/2017/10/how-google-is-us...

exactly. no other SV co is as scientific minded as Alphabet. People focus on the fact that they make money from ads, but nobody looks at what they have done with that money. No other company has done so much for education in my opinion indirectly.( Youtube, Google Earth , Google itself). Google Classroom is also there but I am talking about education in general. As a student, if I had to wish which company survived among the top 5 if 4 had to die tomorrow, then most people in their right mind would vote for Alphabet. even haters.

The list of mostly dudes seen as programmers worth worship makes me sad. Most great software is usually written implicitly or explicitly as a team.

Here’s a name for a start. Jessie Frazelle. Very influencial in the Kubernetes project. I’d love to see a list of “industry changing” female programmers. Mostly because it’s 2018 and we need to change the perceptions that “great software can only be written by men”

So many chiefs.

Jeff Dean is the AI Chief, and Fei Fei Li is the Chief Scientist of AI/ML

Not sure what the distinction is, but good for both of them.

Presumably there are some googlers around here that can explain the difference in detail,but Fei Fei Li is R&D in Google Cloud whereas Jeff Dean is taking the organisational role as head of AI. So much more of the organisation will be involved - some product development, marketing, operations etc. as well a R&D.

off topic: how much do you think a guy like Jeff Dean is worth? 18-19 years at Google, pre-IPO and a super-programmer. Keeping in mind that Google paid Anthony Levandowski $120 Mil and Uber way more than that (technically Uber bought his company and maybe a hard drive :))

More than $500 Mil?

Jeff is priceless.

Levandowski got $120M through acquisition, not salary as an engineer, did he not?

Anyway, there was an old article about top salaries at Google that speculated Jeff making $3M/yr [1].

Jeff also angel invested in Teleport, which got acquired [2]. And his wife and him also have a philanthropy foundation that recently donated $1M to MIT [3] (to promote diversity).

Any more recent employee stock options he gained and past IPO returns could obviously vary significantly depending on percentages, length of time, etc. But I'd be surprised if it was as high as $500M...

[1] https://www.itworld.com/article/2693353/man-or-myth-the-3-mi...

[2] https://www.crunchbase.com/organization/teleport-2

[3] https://www.insidephilanthropy.com/home/2016/8/10/one-of-goo...

Jeff was between employee 20 and 30, IIRC, so it's hard to imagine him not having made many millions with the stock. On the other hand, he's the guy who kept his Volvo for well over a decade.

We're all guessing, but his net-worth probably didn't come from his salary. He joined Google 1-3 years after Google was founded so we have to assume that he got a lot of stock options, especially given his talent. When Google IPO-ed in 2004 it had a market cap of $23 Billion. Today is 30 times that much.

Of course he might have sold shares, but then he might have gotten super bonuses for a lot of his projects. Maybe he asked to leave and Sergey and Larry gave him a new title and lot of money. Sometime a small tweak makes a Google-sized company billions or tens of billions. They are generous with a tiny % of the sum.

https://www.cnbc.com/2017/04/03/waymos-uber-lawsuit-reveals-...

"Embattled engineer Anthony Levandowski collected $120 million from Google, despite involvement with at least one start-up that would ultimately compete with the company, according to new legal filings. "

Levandowski got $120m in pay from Google. His acquistion was when Uber acquired Otto

Google is worth $700B and Jeff Dean was employee ~20, a senior engineer and manager/VP. >0.1% equity for someone of that stature is not unlikely -- mid-level engineers at early-stage startups get about 0.1% equity.

I'm curious what this means practically. What will he be doing all day? Is it still research or code writing, or at this level is he focused on managing people and projects at a high level? If the latter, what does that even mean?

I believe was worthy. Jeff Dean deserves it.

Jeff Dean writes directly in binary and provides the source code as documentation for other developers.

https://www.quora.com/What-are-all-the-Jeff-Dean-facts

My favorite: Compilers don't warn Jeff Dean. Jeff Dean warns compilers.

LOL, Jeff Dean's pin number is the last 4 digits of pi.

> Cricket matches used to take 5 days, until Jeff optimized them.

Unfortunately some matches still take 5 days. Test matches!

Rumors say he writes code in zcat to go faster.

I prefer a Hollywood Director as my AI Chief

For what it's worth, there is no "AI" division at Google. That term just doesn't enjoy any currency inside the company.

Brain, DeepMind and MI are mostly researchy. I know some people working on applied deep learning that got shuffled multiple times between PAs in the last year - it seems to be the solution for this exact problem.

What is MI?..

Machine Intelligence

A contradiction in terms, or an oxymoron?

MI - machine intelligence. Actually, it sounds like the article actually means MI when it says AI. John Giannandrea mentioned in the article used to be a head of "search & MI". So it makes sense that Jeff Dean would become a head of MI.

This is a cool name, I think people should use it more widely instead of AI.

Reply on news.ycombinator.com