Generally AI: Time to Travel

Transcript

Roland Meertens: Welcome to Generally AI, where today the topic is "Time to Travel". And the first fun fact for today is about time zones, because we both had to travel to QCon San Francisco. And I was looking at the time zone database because a couple of weeks ago we went back from summer time to winter time where both your country and my country do this at a different point in time.

Anthony Alford: Isn't that great?

Roland Meertens: Yes. But there are weirder things besides the fact that different continents change from summer to winter time at different moments in time. So I looked at the daylight saving time transitions because most countries go back and forth one hour, but there are two odd cases. One is the Troll station in Antarctica, which is a research station. They have a two-hour jump.

Anthony Alford: Oh.

Roland Meertens: Yes. And I assume that is…I think it's a Norwegian research station. So I assume it's either to be in sync with Norwegian time or with other stations around them.

Anthony Alford: Could be.

Roland Meertens: But six people, they're creating an anomaly in a database. The other fun spot is Lord Howe Island, which puts the clock back and forth 30 minutes.

Anthony Alford: Interesting. Where is that?

Roland Meertens: I'm assuming somewhere around Australia. I don't know exactly where it is.

Anthony Alford: Because there is an Australian time zone that's like 45 minutes off.

Roland Meertens: Oh, I only found this one, which is 30 minutes.

Anthony Alford: Maybe.

Roland Meertens: Also in this case there are 382 people living there and they have a rule that they can have up to 400 tourists at a time to not get too many tourists compared to normal population. And I think this is yet another weird thing developers have to keep in mind, especially with their cron jobs.

Anthony Alford: I work with a lot of people in India and they are in one of those time zones that's a half hour off. So right now they're 10 and a half hours different from the East Coast US.

Roland Meertens: But they don't switch their times.

Anthony Alford: They do not switch their times.

Roland Meertens: No. They just go-

Anthony Alford: Because they're very close to the equator and so they don't need that daylight savings.

Roland Meertens: Yes.

Anthony Alford: And the reason they're a half hour off is because the country is almost the width of a time zone at its widest part. So they're able to put the whole country on one time zone, but it wouldn't work if they aligned it with the hour because one part of the country would be really off.

Roland Meertens: Yes.

Anthony Alford: So they shifted it about half an hour to be a compromise.

Roland Meertens: Interesting, compromise. I also still hate the concept of daylight saving times.

Anthony Alford: Every year people complain about it, but somehow we never seem to get off it. But it's at the state level, right? Arizona doesn't do it.

Roland Meertens: Wait. Different states do or don't?

Anthony Alford: And not only different states can decide, but different portions of the state could be in different time zones or choose not to do it. Hey, welcome to America.

Roland Meertens: Absolute mess.

Anthony Alford: It's like the Holy Roman Empire in some ways.

Roland Meertens: I absolutely love it.

Okay, welcome to Generally AI, an InfoQ podcast where today the topic is "Time to Travel". We are recording a special episode from QCon San Francisco.

Anthony Alford: Yes, I think you kind of buried that lead, but this is really exciting. It's my first one.

Roland Meertens: Yes. Anyways, my name is Roland Meertens. I'm joined by Anthony Alford.

Anthony Alford: Welcome, glad to be here.

Roland Meertens: And we are meeting today for the first time in person.

Anthony Alford: That's right.

Roland Meertens: So that's-

Anthony Alford: You're a lot taller than I expected.

Roland Meertens: Oh, you're also taller than I expected. Turns out it's a podcast recorded by two very tall guys. You didn't know that when listening to this all this time.

Anthony Alford: You would think my voice would be deeper.

Roland Meertens: Yes. Anyways, we both had to travel and change our clocks to get here. And this is why we are talking about software and machine learning topics, which are either related to time or to traveling.

Anthony Alford: That's right.

Roland Meertens: I want to talk about three fun facts related to temporal misunderstandings in data science or machine learning.

Anthony Alford: Oh, very cool.

Google Flu Has Flown [04:16]

Roland Meertens: So I was thinking what interesting temporal machine learning is there, or what's interesting temporal data science is there and where are there mistakes? And I actually thought of two ones I know personally and one I don't, I've never worked with.

Anthony Alford: Two mistakes you made personally or two that you're familiar with?

Roland Meertens: Well, one which influenced me in that I had to go around in a self-driving car to collect more data. So the first thing which I think is interesting is the Google Flu project.

Anthony Alford: F-L-U-E?

Roland Meertens: F-L-U. So it's the flu. When you get sick, you get flu. And it is yet another project by Google which was retired.

Anthony Alford: Okay. The long list. The honor roll.

Roland Meertens: Yes. Well, so I think the idea was really good in that Google thought we can probably see patterns in flu data based on what people are searching in their favorite search engine.

Anthony Alford: Seems reasonable.

Roland Meertens: Yes. So if you see a lot of people start querying best nose wipes, that probably indicates that somewhere someone has the flu, unless they are really passionate about-

Anthony Alford: People are searching for "flu symptoms" or "I have a stomach ache".

Roland Meertens: Yes. So they never told us what keywords they used. I'm assuming it's a very large collection. Maybe someone just analyzed 10 million queries and said these actually correlate to flu. And the hope they had was that they could predict the flu patterns in the US faster than just asking doctors how many people visited you with flu? They retired the project. And I think it's because they didn't always make the best estimates. And I think that the problem here is that correlation isn't always causation.

Anthony Alford: That's what they say.

Roland Meertens: That's what they say. But I think that things which can go wrong here is that…so you know what people Google, you do not know why people Google it. So I think the biggest problem they had was that there was an off-season flu epidemic, which they didn't account for because it's just so different from the baseline. But then I also think that they probably had a lot of false hits whenever maybe a TV show would air a hospital series or a new hospital series would go live.

Anthony Alford: There's a lot of noise in this data.

Roland Meertens: Yes. People start Googling for the symptoms you see there and then you get a lot of noise. And that's why I say you don't know why people are Googling something. Could be because they have it or could be because they see it on a TV show.

Anthony Alford: Well now they can integrate the LLM and the LLM can say, "well why do you want to know?".

Roland Meertens: Yes, yes, but would you want to have a search engine which asks you all the time, why do you-

Anthony Alford: "Why are you asking that?"

Roland Meertens: It would be nice if it just says at 3:00 AM, "Hey Roland, you know that penguins can't fly. You don't need to care about this". Yes. Anyway, so that is something which I think is interesting where temporal patterns really matter and it can be that, I don't know, maybe the one specific, nose-wipe producer had a new commercial and there goes your signal.

Anthony Alford: That's interesting.

Roland Meertens: Yes. I was reading a couple of papers about this and I think one of the articles had quite a good point where they said if they would've integrated all the data, so both search queries and actual signals from doctors, like actual lab results, they could have probably built a better model.

Anthony Alford: That makes sense.

Roland Meertens: Yes, but right now it was really like a versus thing where they either wanted to have completely one party only going by lab results and thus has a three-week delay and the other party only goes by search queries.

Anthony Alford: Well, they have the search query data right there in real time, whereas like you said, the lab stuff, there's some latency and they have to collect it somehow.

Roland Meertens: Yes.

Anthony Alford: So that makes sense.

Roland Meertens: I think it's a really cool project. I would love it to still be alive, but no, it's gone.

Anthony Alford: Yes.

Dynamic Pricing and Natural Intelligence [08:17]

Roland Meertens: The other thing which I wanted to talk about was dynamic pricing, especially for unpredicted events. So there are many cases where, for example, a snowstorm starts, which wasn't predicted by the ride hailer, and of course people start ordering more taxis because they just don't want to be outside or they start ordering more food.

Anthony Alford: Right.

Roland Meertens: This is something that tends to hurt startups where they didn't predict that.

Anthony Alford: Yes, well now that they're in the cloud, they can auto-scale.

Roland Meertens: Yes. I do always think it's kind of interesting if there is some hurricane and you still see an Amazon delivery driver, just a prediction for when the package-

Anthony Alford: As someone who lives in the South, the southern part of the US, the Southeast. So we get both hurricanes - and you have that behavior where people go and buy all the bottled water and generators, but when we get a thread of snow, people will go and buy up all the bread, milk and eggs. The stores will just be out of those items. I don't know why, nobody knows why.

Roland Meertens: Yes.

Anthony Alford: But it's something that happens.

Roland Meertens: Yes. With the pandemic, people started buying the weirdest stuff as an emergency. But yes, the one story which I have here is that I had this friend who was working at a bus company which would rent out buses, rent them out for a preset price for specific dates. So you could go to the website and say, I want to have a bus from here to here for a team event.

Anthony Alford: Sure.

Roland Meertens: And then they would get you a bus, but they would rent it from other companies because they are a platform. That's what companies do nowadays. The problem was that they set the prices based on historic usage in Germany and then they expanded to other countries.

Anthony Alford: Interesting.

Roland Meertens: For example, the Netherlands and other countries. And the problem was that although Germany and the Netherlands are often confused by other people, they tend to have different holiday days and different events which attract a lot of people.

So there were moments where all over the countries, all the buses were already sold out because there was maybe a major event in Amsterdam. So everybody from all over the country wanted to go there and they kept selling their buses cheaply because in Germany nobody wanted to go to Amsterdam that day.

Anthony Alford: Sounds like the model was over-fit.

Roland Meertens: Yes, but I think it's interesting. It's one of those things where you just think, oh, we have a model for a similar country, let's just deploy it and see what happens. And things can go wrong.

Anthony Alford: Seasonality is one of those things that - you can have a time series prediction.

Roland Meertens: Yes.

Anthony Alford: Mostly it's going along, it's fairly predictable, maybe a little bit of linear growth and then something seasonal…just kaboom.

Roland Meertens: Yes. But there's just massive differences. So I know that at previous QCons there have been some talks from Uber because they often go to a new market and they need to get their pricing right from the get-go, which can be extremely hard. If you decide to launch in China and you don't predict a Chinese New Year sort of thing, then maybe you are underpricing very quickly.

Anthony Alford: Probably helps to have some local experts.

Roland Meertens: Yes. I also don't really know how to solve it in this case because I guess you need people in operations to be very close to your data science team.

Anthony Alford: Yes.

Roland Meertens: Then again, if you are a data scientist and someone comes to you and says, your model was wrong because it didn't predict this national holiday, it's not really data science, it's just common sense.

Anthony Alford: You need to have a little bit of natural intelligence.

I Want to Ride My E-Scooter [11:48]

Roland Meertens: I guess. I disentangle data science and common sense now. The last story which I wanted to share is, as I said, I had myself when I was working in Munich at a self-driving car company. We had a machine learning system which had the following classes: car, bike, pedestrian. And then one day Germany made it legal to drive e-scooters on the street.

Anthony Alford: Okay.

Roland Meertens: And this was really from one day to the next, that five scooter companies came into Munich. So yes, it's an introduction of a new object. We had very tightly defined classes.

Anthony Alford: It's a really fast moving pedestrian.

Roland Meertens: Yes. They sped our pedestrians up without giving us a proper warning. And our car didn't respond very well to this. So because of this seasonality change or this introduction of a new concept, I had to drive around with a car and collect data as fast as possible to get it annotated quickly and get our model-

Anthony Alford: Really?

Roland Meertens: Yes.

Anthony Alford: Okay. I assume you need to characterize some parameters of this e-scooter person…

Roland Meertens: It's a mix between... So this was a traditional automotive stack. So we had separated the object detection, which was what I was working on, and the prediction stack.

Anthony Alford: Gotcha.

Roland Meertens: So that was what someone else was working on.

Anthony Alford: So it's like the dynamics for your Kalman filter or whatever?

Roland Meertens: Yes. Yes, yes. So I guess that they had to adapt the parameters themselves in their system.

Anthony Alford: Okay.

Roland Meertens: But yes, it is fun when new objects get introduced, you have a very tightly defined system and you need to deal with them.

Anthony Alford: That's interesting.

Roland Meertens: Yes.

Anthony Alford: So you break your enumeration now and you've added a new class, you gotta retrain everything.

Roland Meertens: Yes. But generally if you are making all your objects very tight, then things keep breaking over time, very often people decide that they want to have a new object annotated. But yes, can be interesting in terms of scalability and in terms of time.

California Here I Come [14:16]

Anthony Alford: Hey, so we're back and at the beginning of the show we were talking about time zones, and you probably know that time zones came about because of railroads about 150 years ago, more or less. Because to make the trains run on time, or to appear to run on time, everyone has to agree what time it is.

Roland Meertens: Yes.

Anthony Alford: So instead of setting your clock by the sun, now you set your clock the same as everyone else in your area. So I'm going to talk about railroads, but I'm going to tie that to San Francisco, which is where we are.

We're actually here right at the very tip of San Francisco at Embarcadero near the ferry terminal. But a little south of here between Menlo Park and Palo Alto, there's a little private university you might've heard of. It's called Leland Stanford Jr. University.

Roland Meertens: I heard of something called Stanford.

Anthony Alford: Yes, its official name is Leland Stanford Jr. University. Leland Stanford Sr. was the founder of the school. He named it for his son who died very young. And the elder Stanford was the governor of California from 1862 to 1863.

He's probably best known to history as one of the key investors of the Central Pacific Railroad. And this railroad is half of my story. The other half of the story is the Union Pacific Railroad. But it's really one story, the story of the first transcontinental railroad that joined California and the Pacific Coast with the rest of the country.

And by the way, there's an excellent book about this project, which I have relied on heavily. It's called Nothing Like It in the World by Stephen Ambrose. And the reason for that title is: this was the first transcontinental railroad.

Roland Meertens: Yes.

Anthony Alford: There was nothing else like it. There were of course a lot of railroads. Anyway, the San Francisco connection, as you might guess: the western terminus of this railroad, it was not quite in San Francisco. It was across the bay in Oakland.

Roland Meertens: Yes.

Anthony Alford: Close enough.

Roland Meertens: Yes.

Anthony Alford: And the railroad was completed in 1869, and that was the first opportunity for passengers to make a contiguous rail journey from the East Coast of the US to the West Coast.

Roland Meertens: To almost the West Coast.

Anthony Alford: To the water.

Roland Meertens: To the Bay.

One Way by Land, Two Ways by Sea [16:34]

Anthony Alford: Yes. Of course, there were a lot of railroads on the eastern side of the US, on the east of the Mississippi. Anyway, we're getting ahead of ourselves. Before 1869, if you wanted to travel from that part of the US to the West Coast, here were your choices.

Number one, overland. This is difficult and dangerous. A lot of people did do this of course, and there weren't really roads as we know them now, but there were a lot of commonly used trails. Maybe you've heard of a game called Oregon Trail.

Roland Meertens: Yes. "You die of dysentery".

Anthony Alford: Right. Well, so you have an idea of some of the dangers involved, and it took a very long time. It was a long way, about 2,000 miles. There's not one, but two formidable mountain ranges.

Roland Meertens: Yes.

Anthony Alford: I don't know what they are in meters, but it's like 14,000 feet or 10 to 14,000 feet. And in between them there's a desert.

Roland Meertens: Yes.

Anthony Alford: So there's a lot of opportunities for disaster. Now, if you didn't want to walk or ride across land, you could go by sea. And there are two options here.

One, you could take a boat all the way around, which meant going all the way to the southern tip of South America and then back north. So going overland could take you up to six months. This trip by sea, maybe four to six months depending on wind at first, but there's steam ships. By the 1860s it was steam ships. But it was still quite a long trip.

But there is in between North and South America, a very narrow strip of land and there was no canal then, but it was possible to take a boat to Panama or Nicaragua and then cross over to the other side and get a ride on another boat.

Roland Meertens: Yes.

Anthony Alford: And that would only take you a couple of months. But a lot of times people would get sick. The eastern coast of Nicaragua is called the Mosquito Coast. So you can imagine that.

Roland Meertens: Yes.

Anthony Alford: And in fact, when the Panama Canal was built, a lot of people got sick and died of fevers working on that. So it was a great way to get sick and it's not a great choice for bulk cargo. If you're going to send cargo, you're going on the boat all the way around.

Interestingly, one of the ways to get across Nicaragua, there's a very large lake and this guy named Commodore Vanderbilt owned some steamships on that lake. So there's another university namesake.

In fact, really if you think about it, Stanford is the Vanderbilt of the West, you might say. Anyway, you might wonder now, such a pain to get there. Why was anybody in a rush to get to California? It was a gold rush, right? In 1849, gold was discovered not far from San Francisco.

Roland Meertens: Where is the gold now?

Anthony Alford: I don't know. I think it was all turned into jewelry and money.

Roland Meertens: So there's nothing left?

Anthony Alford: Well, we're still here.

Roland Meertens: Can I go digging outside?

Anthony Alford: We've started digging up silicon instead of gold.

Roland Meertens: Yes.

Anthony Alford: But there were a lot of people that were headed west to make their fortune. And as the saying goes, the real winners here were the people who were selling the shovels and other stuff. So Stanford was one of those people selling the shovels. He was a merchant. I think he actually sold groceries.

Roland Meertens: He was literally selling shovels.

Anthony Alford: I think he was selling other things, but he made a lot of money selling things. But the shovels and all the stuff were made in the east, so people had to get that stuff over there too, right? So clearly everybody wanted a railroad to get the stuff and the people there quicker.

Roland Meertens: Yes.

Railroads - The Original Tech Sector [20:07]

Now, I'm a modern-day American, and as you pointed out, we're living the GTA lifestyle here, so we go everywhere by car. So it might surprise you and some of our listeners to know that here in the USA we used to be quite keen on railroads. And in the mid-eighteenth century it was what today we would consider an exponential growth industry.

So in that book by Stephen Ambrose, he says the distance of track more than doubled in each decade. In 1834, there were only 762 miles. In 1844, it was up to 4,300 miles. By 1854, it was 15,675 miles.

Roland Meertens: So they saw the hockey stick curve.

Anthony Alford: Right. So the eastern part of the country was being crisscrossed by railroads by this time. So starting with the gold rush, this really kicked off a lot of buzz trying to build this railroad from the eastern part of the US to California. And there was a scrappy young backwoods lawyer in particular pushing it. It was a man named Abraham Lincoln.

In 1859, he was giving a speech in Council Bluffs, Iowa. And he met a young engineer named Grenville Dodge, who was introduced to Lincoln as knowing more about railroads than any two men in the country. Dodge convinced Lincoln that the Transcontinental Railroad should begin right there, actually across the river from Council Bluffs at a place called Omaha, Nebraska.

Roland Meertens: Yes.

Anthony Alford: Now, spoiler alert, Dodge in fact was to become the chief engineer of the Union Pacific. And he was to begin building that railroad from Omaha. Because it turns out in a couple of years, Lincoln was in a position to get the federal government to agree to finance building this railroad.

Roland Meertens: Yes. This Lincoln guy will go places.

Anthony Alford: He did go places. So he actually got some legislation pushed through. But both he and Dodge were distracted by other matters until 1865. And of course, sadly, Lincoln's life was cut short.

But to sum up, the US government agreed to back the construction of the railroad as a joint project by both the Central Pacific and the Union Pacific. And the way they funded it was by giving each of these railroad companies bonds and land grants, and each would be paid according to the distance of track that they laid.

And their tasks were to head out from their respective locations. So the Union Pacific from Omaha, the Central Pacific, they actually started in Sacramento, but they eventually pushed it back to Oakland. So they were going to start out from their starting points and head generally towards Utah, towards Salt Lake City.

Roland Meertens: But there was no plan of where they were going to end. They were just starting-

Anthony Alford: Yes, it was deliberately vaguely worded because number one, it was unclear how fast each was going to progress.

Roland Meertens: Oh, yes.

Anthony Alford: And number two, it was also not 100% clear what was the optimal route, because like I said, there's mountain ranges and deserts. There was a lot of exploring and surveying that had to be done because the railroads were…first of all, there were stipulations on the maximum grade. The power of the engines could only handle a certain steepness. So they had to be really careful about that.

But they generally knew approximately where they would wind up. And it turned out to be near Salt Lake City, which is where KubeCon was last week. We had a lot of people show up like Daniel was at KubeCon last week. Anyway, going off on a tangent.

Now, the Central Pacific, they had the rougher sledding because they start out in Sacramento, it's less than 100 miles and they're in the mountains.And they're some pretty steep mountains. But the Union Pacific, they're on the prairie. It's relatively flat.

Roland Meertens: Yes.

Anthony Alford: And the Central Pacific had the additional disadvantage that all of their stuff had to come from the east by ship.

All the rails, all the engines, all the tools, all the gunpowder. They used gunpowder to blow holes in the mountains to get through.

So you might wonder, what does this all have to do with a tech conference? So let's talk about that. Well, as I said earlier, railroads were the high-tech industry of their day. They had exponential growth. And there's a lot of other facets we might recognize. First of all, these railroads were doing continuous delivery.

Roland Meertens: Yes.

Anthony Alford: And in fact, they were doing continuous delivery so well, they could at their peak lay down 10 miles of track a day. Almost as fast as a person could walk next to the track, they could lay it down. All manual labor.

And they were drinking their own champagne because they had to use the track that they're laying behind them to bring up more supplies. So they're basically bootstrapping everything. Right?

They definitely had the mindset of move fast and break things, or at least move fast. They needed to lay that track as fast as possible because they were kind of in a race against each other.

Roland Meertens: Yes, you get paid per mile.

Anthony Alford: You get paid per mile. You want to be laying that as fast as you can. So at the beginning, they were really struggling to lay 50 miles in a year, but as I said, they got up to 10 miles a day at some points.

Interestingly enough, they also made some deliberate, suboptimal technical decisions that allowed them to build faster with the intention of refactoring later. So we would call this tech debt. For example, on the prairie, there aren't any trees. There's a few trees called cottonwoods. But their wood wasn't really good for making railroad ties, but they used it anyway because it was there.

The alternative was to have more expensive wood shipped or transferred, right? So they used these locally available cottonwood ties knowing they would have to be replaced in just a few years.

Roland Meertens: Yes.

Live-streaming the Golden Spike [26:01]

Anthony Alford: And in the end, as I said, the two lines met near Promontory, Utah, which is near Salt Lake City. It's a place on the north side of the lake. I've actually been there. There's a nice little reenactment of the ceremony that happened when they met. There was a ceremony where they drove the last spike. It was a golden spike. So there's where the gold went.

Roland Meertens: Yes. The railroad.

Anthony Alford: In fact, it was Mr. Leland Stanford Sr. who drove that last spike and they live-streamed it. Here's how. Of course next to the railroad, there was a telegraph line. The telegraph followed the railroad, actually, the telegraph preceded the railroad in a lot of cases. They put a wire on the hammer and on the spike. So when Stanford tapped that hammer on the spike: tap, tap, tap, that signal went out on the telegraph.

Roland Meertens: Oh, nice.

Anthony Alford: And then they drove a locomotive from each end forward until they touched. And that was the railroad: it was open. And then with that, the US had rail service from coast to coast.

Roland Meertens: Nice.

Anthony Alford: And instead of a six-month trip, now it's about a week or then it was a week.

Now, today, as you know, we love our cars and except for a few metros like the Bay Area here where they have the very fine BART rail system, people don't travel by rail much in this country. We actually do ship a lot of freight. As a percentage, I think it's more than other rail enthusiast locations like Europe and Japan.

Roland Meertens: Can you travel from the east coast to the west coast by rail?

Anthony Alford: You can. We actually do still have passenger service. It's slow.

Roland Meertens: How many days?

Anthony Alford: I don't know, but it's probably about the same as it was. It's usually... It's further north from Chicago latitudes across to Seattle or something.

Roland Meertens: Yes, yes.

Anthony Alford: But yes, there's still several passenger rail services in the US. But it's true, for moving people around, we're either flying or driving. In fact, I flew here to San Francisco from the East Coast of the US and it only took about six hours of actual travel instead of six months.

Roland Meertens: It's disappointing that you didn't take the train.

Anthony Alford: I would still be on the train. So almost to wrap up, we love our driving. You probably are familiar with our interstate highway system. This began construction in 1956. It took until 1986 to have a single continuous coast to coast interstate highway. That's Interstate 80.

Roland Meertens: It was later than I would've expected.

Anthony Alford: Well, they did it in pieces, right? So the interstate would be constructed and this state is going to build a strip. Anyway, interstate 80 connects New York with…San Francisco. And in fact, the route, once you get past Omaha, generally follows that of the Transcontinental Railroad. And in fact, the final section of Interstate 80 was built in Utah, right on the edge of Salt Lake City.

So as they say, history does not repeat, but it does rhyme.

Roland Meertens: Yes. Yes, yes.

Anthony Alford: And that's my content for QCon San Francisco Special Edition podcast of Generally AI.

Roland Meertens: Nice. All right. Let's go to words of wisdom. What session did you enjoy the most yesterday or today?

Anthony Alford: Well, today, actually I enjoyed that keynote.

Roland Meertens: Yes, me too.

Anthony Alford: It was about social factors for productive teams.

Roland Meertens: Yes, the title is Not So Hidden Social Drivers Behind the Highest Performing Engineering Teams by Lizzie Matusov.

Anthony Alford: Yes.

Roland Meertens: Extremely interesting. Yesterday I also saw a lot of talks about search and optimizing search.

Anthony Alford: As did I. In fact, I'm covering some of them as news.

Roland Meertens: Nice. Yes. Did you publish the Netflix talk where they showed how they improved the search and personalized it for everybody?

Anthony Alford: Yes, that was right. That was nice.

Roland Meertens: That was an amazing talk. I really enjoyed that.

Anthony Alford: Well, this was a great idea. It was good to finally see you in person.

Roland Meertens: Yes.

Anthony Alford: And share some stroopwafels!

Roland Meertens: Thanks for listening. This was Generally AI, an InfoQ podcast. If you like this podcast, you can give us a rating in your favorite podcast app. I also say that you should just recommend it to friends whenever they ask you what podcast to listen to. Anyways, thank you everybody for listening. Have a nice day.

Anthony Alford: See you.

Mentioned:

• Google Flu
• Nothing Like It in the World
• Golden Spike Site