I remember I was at Microsoft more than a decade ago now and at the time there was a lot of concern about search and how far Bing lagged behind Google in geospatial (maps).
After some initial investment in the area I was at a presentation where one of the higher ups explained that they'd be abandoning their investment because Google Maps would inevitably fall behind crowdsourcing and OpenStreetMap.
Just like Encarta and Wikipedia we were told - once the open source community gets their hands on something there's just no moat from an engineering perspective and once it's crowdsourced there's no moat from a data perspective. You simply can't compete.
Of course it's more than a decade later now and I still use Google Maps, Bing Maps still suck, and the view times I've tried OpenStreetMaps I've found it far behind both.
What's more every company I've worked at since has paid Google for access to their Maps API.
I guess the experience made me skeptical of people proclaiming that someone does or does not have a moat because the community will just eat away at any commercial product.
I've been using osm more and more recently. Google just makes a bunch of frustrating decisions that really pushed me to look elsewhere. Especially in the public transport layer, but more generally in being really bad at deciding when to hide details with no way to override it and say "TELL ME THE NAME OF THIS CROSS STREET DAMNIT THATS THE ONLY REASON I KEEP ZOOMING IN HERE!!!".
> generally in being really bad at deciding when to hide details with no way to override it and say "TELL ME THE NAME OF THIS CROSS STREET DAMNIT THATS THE ONLY REASON I KEEP ZOOMING IN HERE!!!".
Stuff like this is the main reason I end up switching to Apple Maps on the occasions that I do so. Another example is refusing to tell me the number of the upcoming exit I'm taking.
In general I would say Google Maps is still superior to Apple Maps, but between the aforementioned baffling design decisions and Google maps now including ads in destinations search results, I find myself experiencing more and more mental friction whenever I use it.
Google maps is at least getting better about screen real estate. I have an android head unit, and Maps clearly assumed you'd always be using Maps in portrait mode, because the bottom bar would clutter up the bottom of the screen with "local stuff near by you might be interested in" if you weren't actively navigating.
Eventually switched to Waze, which is now also cluttering things up with (basically) ads.
That's funny because when driving in the bay area, inability to get the -name- of the upcoming exit from google maps was driving me nuts! The exit numbers are not listed on the upcoming exit/distance signs on 280.
Both apps seem to get the names of exits consistently wrong in the Bay Area. I don’t care what a map thinks the name should be — I care what the sign says.
The inability to easily get a street name is one of my biggest pet peeves with Apple Maps, it's up there with the generally poor quality of turn-by-turn navigation (at least in the Bay Area).
There is a spot in NYC where zooming in on my iPhone in Apple Maps in satellite view causes the app to crash somewhat reliably. It has been happening for the last few months.
This. My car uses Google Maps for its built-in nav system, and I've spent a lot of time on road trips wondering just what state I was in. It's insane that Google hasn't added something as trivial and important as state borders.
Open source works well when the work is inherently cool and challenging enough to keep people engaged. Linux and Blender are two of the most successful open source projects, and the thing they have in common is that problems they solve are problems engineers enjoy working on.
Mapping intersections is extremely boring in comparison. The sheer quantity of boring work needed to bring open street maps up to the quality of google maps in insurmountable.
LLMs are freaking cool, and that bodes well for their viability as open source projects.
My impression is that open street maps problem is not the map quality. In areas I have used it, it often has details (e.g. small hiking paths, presence of bike lanes) that google maps doesn't have.
The issue is search. Searching for things that you don't know precisely (music bars in this area). This type of data/processing on top of the geospatial was always subpar and very hit or miss in my experience.
That’s not my experience. I work in downtown Minneapolis and open street maps is missing basic things like entrances to public parking lots. Open street maps has a problem if it can’t get details right in population dense areas.
It really depends, but in germany (has a large OSM community) OSM has so much better quality & detail for almost everything except buisnesses. It suffers from poor search, routing that doesn't take traffic jams or roadworks into account and a lack of high quality apps and thus only "nerds" use it instead of Google Maps or others.
That's ... probably not the best example, given the fact that there are a shedload of open-source databases of various types that have forced major commercial vendors like MSFT and ORCL into a corner. ORCL's moat is that they have a large portfolio of random solutions, are incumbent at a lot of organizations where switching costs are very high, and they have an exceptionally aggressive sales organization that doesn't seem to worry too much about legalities.
This is an instructive error. From my perspective, there was plenty of evidence even 15 years ago that community efforts (crowd-sourcing, OSS) only win sometimes, on the relevant timeframes.
So the “higher ups” were using too coarse a heuristic or maybe had some other pretty severe error in their reasoning.
The right approach here is to do a more detailed analysis. A crude start: the community approach wins when the MVP can be built by 1-10 people and then find a market where 0.01% of the users can sufficiently maintain it.[1]
Wikipedia’s a questionable comparison point, because it’s such an extraordinary outlier success. Though a sufficiently detailed model could account for it.
In terms of data, OSM is so far ahead of Google maps in my experience. The rendering is much better too. What's not there is obvious and easy to use tooling that anyone can interact with. I mean, there might be, but I don't know about it.
I don't have google maps on my phone at all unless I visit in the browser, and I use OSM through Magic Earth. I wouldn't go back, but it is a huge pain and sometimes I do have to just open google maps in a browser window. It doesn't usually have hours of operation, doesn't usually have links to websites. It can't find businesses by name easily (it often seems to require the exact name to by typed in), and it definitely can't find businesses by service (searching for "sandwiches" will not show you a list of local sandwich shops, it will do something like teleport you to a street called "Sandwiches" in Ireland). And even if I have the exact address, I will still sometimes end of thousands of miles away or with no hits because the street name was written differently. Honestly, it's of very little use to me because it can rarely take me to a new place.
People in the real world care about things like hours of operation. Google makes it really easy for businesses to keep them up to date on things like holiday closures. OSM makes it a nightmare.
How do they make it a nightmare? Are we sure it's not just that 96% of business owners use Google maps or maybe Apple maps and don't even know what OpenStreetMaps exists. I think this is more about network effects then anything. If they really want to break googles geo spacial business data monopoly.
I think if Apple/Microsoft/OSM should band together and make a simple tool for business owners that can update your details on Google, Bing, Apple Maps, and osm simultaneously. Although I am not sure if Google exposes that through apis or not.
I am very sure that OSM does not get this information because they make it hard for businesses to give it. I know this because figuring out how to get that information published was my job a few years ago.
Specifically I was a developer for a company whose job was to update business information in Google Maps, Apple, Facebook and so on. We'd get the data from companies like Cheesecake Factory, Walmart and Trader Joe's, then we would update all of the sites for them.
All of the sites have some sort of API or upload mechanism that makes it easy to do things like publish phone numbers, hours of operation, hours for specific departments and so on. All of them were happy to let us automate it. All were happy to accept data based on street addresses.
I tried to make it work for OSM. It was a disaster. I have an address. Google et al understand that a street often has multiple names. If the address I was given named the street something else, Google took care of understanding that route 33 is also such and so street and they accepted the data. If I said that there was a restaurant inside of a mall, Google didn't insist that I know more than that. If I was publishing holiday hours, Google accepted us as the default authority. (And gave ways of resolving it if someone else disagreed.)
OSM did NONE of that. It was all roadblocks. If I didn't have the One True Name that OSM in its wisdom determined was right, good luck matching on address. If I couldn't provide OSM with the outline of the restaurant on the floor plan, OSM had no way to accept that there was a restaurant in the mall. If a random OSM contributor had gone to the location and posted store hours, OSM refused to accept my claim of its reduced hours on Christmas Day. And so on.
All of the brands that I named and more don't publish to OSM for one reason, and one reason only. OSM make it impossible for businesses to work with them in any useful way to provide that information. And therefore OSM is not on the list of sites that that data gets published to.
In short, if it isn't perfect, OSM doesn't want your data. And the data off of a spreadsheet some business uses to manage this stuff usually is nothing like perfect. I respect how much work went into getting OSM just right. But they made it impossible for real businesses to work with them, and so they don't get that business data.
For starter, as a business owner, how do you claim full ownership of a given business on OSM?
What prevents a nasty competitor from making daily false updates to your opening hours?
If you're a verified business owner in a non-collaborative platform, you can update your holiday hours/one-off closure with a simple edit on that platform's business management page/API. How is OSM even in same category as Apple maps/bing/Google maps?
A nice UI is completely and utterly useless for a business attempting to create an automated workflow from a spreadsheet for things like business hour updates and letting map publishers know when new stores are going to open.
So yeah, OSM is a nightmare for businesses to deal with. And unless that changes, its access to business information that people expect will remain severely limited.
Is there a recommendation for OSM on mobile? IIRC they don't have an official app.
Also looking at their bike routing - it gives me an idea. Road should be rated on whether they have a dedicated bike lane and on the danger of riding on said road at particular times of day. I just input a src/dest and it gave me a really busy road with tons of "paperboy" level risky side roads on it. I would never want someone to take that route at 5pm on a weekday.
OSM is fundamentally just a DB for place locations and geometries. Directions use routing engines which choose roads and paths between locations based on constraints. The main landing page for OSM lets you choose between OSM, Grasshopper, and the Valhalla routing engines.
To figure out why directions are bad you need to see which criteria the routing engine is using to create the route and decide either to change the constraints used to generate the bike route or what added data you need to place on the streets for the routing engine to avoid/prefer certain streets.
Does this sound like an opaque nightmare? Yes. That's why very few people use it. Apple has been doing some great work doing mapping and adding it into the OSM DB, which they use for their own maps, but they have their own proprietary routing system for directions. If you're looking for a good app to use just OSM data, I use OSMAnd for Android. I still prefer Google Maps because their routing and geocoding tend to be much better for urban areas but for hikes and country bike rides, OSM tends to outperform GMaps.
OSM is a database of map data (streets/buildings/etc), so satellite and street view imagery is outside of its scope. Individual map applications that use OSM data might also support satellite imagery (and some do, like OSMAnd).
> Of course it's more than a decade later now and I still use Google Maps, Bing Maps still suck, and the view times I've tried OpenStreetMaps I've found it far behind both.
The sheer size of the OSM project is staggering. Putting it next to Wikipedia, where missing content at some point wouldn't cause much fuss, makes it a bad example.
Besides that, your limited knowledge of the popularity of OSM gives you a wrong picture. OSM is already the base for popular businesses. Like Strava for example. TomTom is on board with it. Meta for longer with their AI tool, same as Microsoft. In some regions of the world where the community is very active, it IS better than Google Maps. Germany for example where I live. In many regions of the world, it is the superior map model for cycling or nature activities in general. Sometimes less civilised areas of the world have better coverage too because Google doesn't care about those regions. See parts of Africa or weird countries like North Korea.
One should also not forget the Humanitarian OpenStreetMap Team which provides humanitarian mapping in areas Google didn't care. You can help out too. It's quite easy: https://www.hotosm.org/
> What's more every company I've worked at since has paid Google for access to their Maps API.
Many others have switched away after google lifted their prices.
They'll lose the race here too.
A simple donation of up-to-date world satellite imaginary would already be enough for an even faster grow.
I think ex YU states and former Soviet bloc also really shine in OSM, as well as areas along PRC border where regime forces map jitter (see HK/PRC border road junctions for example)
The problem with a lot of open source is the long term issue.
The people doing many of these projects often want the short term kudos, upvotes, or research articles. They may iterate fast, and do all kinds of neat advancements, except in a month they'll move to the next "cool" project.
Unfortunately, with a lot of open source projects, they don't want to deal with the legalese, the customer specific integration, your annoying legacy system, the customer support and maintenance, or your weird plethora of high-risk data types (medical industry I'm looking at you)
Not sure what the Wikipedia reference is, since how many people use any form of encyclopedia other than crowdsourced Wikipedia?
However, to note, there are some examples of successful long term open source. Blender for example being a relatively strong competitor for 3D modeling (although Maya still tends to be industry dominant).
Agreed, even the best open source projects, like Linux or Firefox, in their wonderful success, didn't render proprietary competition unable to have there piece of the market share.
And even in markets with very dominant free offers like video consumption, programming languages or VCS, you can still make tons of money by providing a service around it. E.G: github, netflix, etc.
OpenAI has a good product, a good team, a good brand and a good moving speed.
The higher up failed to see the difference in "users", as well as use cases.
In Wikipedia, the user is same as the content creator: the general public, with a subset of it contributing to the Wikipedia content.
In OpenStreetMaps, one category of users are also creators: general public needs a "map" product, and a subset of them like contributing to the content.
But there's another category of users: businesses, who keep their hours/contact/reviews updated. OpenStreetMap doesn't have a nice UX for them.
As for use cases: underlying map data sure, but one needs strong navigation features, "turn right after the Starbucks", up-to-date traffic data.
This all makes it so different from Wikipedia vs Encarta.
This sounds right to me and was similar to my reaction. The doubt I had reading this piece is that GPT4 is so substantially better than GPT3 on most general tasks that I feel silly using GPT3 even if it could potentially be sufficient.
Won't any company that can stay a couple years ahead of open source for something this important will be dominant as long as it can do this?
Can an open source community fine tuning on top of a smaller model consistently surpass a much larger model for the long tail of questions?
Privacy is one persistent advantage of open source, especially if we think companies are too scared of model weights leaking to let people run models locally. But copyright licenses give companies a way to protect their models for many use cases, so companies like Google could let people run models locally for privacy and still have a moat, if that's what users want, and anyway most users will prefer running things in the cloud for better speed and to not have to store gigabytes of data on their devices, no?
This is an excellent point. I think the memo is making a different kind of case though - it's saying that large multipurpose models don't matter because people already have the ability to get better performance on the problems they actually care about from isolated training. It's kind of a PC-vs-datacenter argument, or, to bring it back to Maps, it'd be like saying mapping the world is pointless because what interests people is only their neighborhood.
I don't buy this for Maps, but it's worth highlighting that this isn't the usual "community supported stuff will eat commercial stuff once it gets to critical mass" type of argument.
Google Maps 3D view is unmatched compared to anything open source has to offer.
Let alone the panning and zooming, there is no open source solution which is capable of doing it with such a correctness, even if we ignore Google's superb "satellite" imagery with its 3D conversion. I have no access to Apple Maps, so I can't compare (DuckDuckGo does not offer Apple's 3D view).
Google maps isn't so good because google is good* but because google feeds their maps with data from their users, which is a huge privacy concern that most people simply don't care about.
I use Apple's notably inferior maps because they're not feeding my data straight into their map and navigation products. It's a tradeoff most wouldn't be willing to make, but that tradeoff is why their maps are better.
It boils down to out of date maps are worse than worthless and google has a scheme to keep theirs up to date. It's a huge maintenance problem...unless your users are also the product.
So maps might be a bad comparison to ML/AI development.
*Google using their user data can be interpreted as google being good at it, sure.
As an aside, I stopped using Google maps/waze because I got the distinct impression I was being used as a guinea pig to find new routes during the awful commute I used to have. I would deliberately kill the app when I went to use a shortcut I knew about so that the horde wouldn't also find it via those tools.
I stopped using Google Maps in my car with CarPlay, because the map would lag by about 5 seconds to reality, which is really bad at say 55 mph in a place where you’re not familiar.
Been using Apple Maps now for six months, and very happy with it. No lag, and very useful directions like “turn left at the second stop light from here”.
OSM is quite popular through commercial providers, mainly Mapbox. Why you're not using it daily is because there's no concentrated effort to make a consumer-friendly product from it, like Wikipedia mostly is for Encyclopedia. Too early to tell what will be the case for LLM.
Open source will never defeat a company in areas where the work is very, very boring and you have to pay someone to do the grunt work. The last 20% of most tasks are extremely boring so things like data quality can only be accomplished through paid labor.
This isn't an apt comparison. Maps need to be persistently accurate and constantly updated regardless of community involvement, AI just has to be somewhat applicable to the paid version (which, given its stochastic nature, the open source alternatives are close enough). Microsoft obviously misunderstood the needs of maps at the time and made the wrong conclusion. The lack of moat for AI is closer to the Encarta/Wikipedia scenario than the maps scenario.
I think the difference is that Maps is a product and its hard to copy a whole product and make it good without someone driving the vision. But a model is just a model, in terms of lines of code they aren't even that large. Sure the ideas behind the are complicated and take a lot of thought to come up with, but just replicating it or iterating it is obviously not the challenging based on recent developments.
Just anacdotally, I see OSM mentioned a lot, guides for contributing, use in HomeLab and Raspberry Pi articles-- haven't check it out myself in a long time, but I wouldn't be surprised if it's continued growth really is inevitable, or even has a cumulative snowball-ball component
OSM's main problem is that it has no open sourced satelite imagery dataset to display, they're only using borrowed data to build its vector maps on. It just doesn't exist. Until that becomes a thing it'll stay a second rate map app for the average person, unfortunately.
It's the only map anyone can actually integrate into anything without an api key and a wallet with a wad of greens in it, so that keeps it relevant for now. Maybe if/when Starship lowers cost to orbit, then we'll see non-profit funded satellites that can source that dataset and keep it up to date.
Do you happen to know why there isn't any U.S. Government satellite imagery? I understand the really high-resolution stuff is probably from spysats and so classified, but anything else should be public domain, no?
Everything under NASA's and NOAA's purview is public domain. High resolution stuff is left to commercial and secret applications. Some states also have high res aerial photography. This was notably obvious in the early days of gmaps when the whole US was Landsat only with aerial for just Massachusetts.
What if instead of Microsoft abandoning their investment they'd invested directly in OpenStreetMap? Because that seems more analogous to the course of action the article is recommending.
Crowdsource is significantly different from open source.
Open source is Linux winning because you don't need to pay Microsoft, anyone can fork, Oracle/IBM and Microsoft's enemies putting developers to make it better and so on. Today .NET runs on Linux.
Crowdsource is the usual bs that either through incentives (like crypto) or by heart, people will contribute to free stuff. It doesn't have the openness, liberty or economic incentives open source has.
And Google has lots of crowdsourced data on Maps, I know lots of people who loves to be a guide there.
Data is still valuable and you can build a moat with it. But this discussion isn't about data, it's about models.
A better analogy would be paywalled
general-purpose programming languages, where any access to running code is restricted. Such a programming language would get virtually no mindshare.
This Google employee is just saying, let's not make that mistake.
Even if Google fired all AI researchers tomorrow and just used open source models going forward, they could still build killer products on them due to their data moat. That's the takeaway.
After some initial investment in the area I was at a presentation where one of the higher ups explained that they'd be abandoning their investment because Google Maps would inevitably fall behind crowdsourcing and OpenStreetMap.
Just like Encarta and Wikipedia we were told - once the open source community gets their hands on something there's just no moat from an engineering perspective and once it's crowdsourced there's no moat from a data perspective. You simply can't compete.
Of course it's more than a decade later now and I still use Google Maps, Bing Maps still suck, and the view times I've tried OpenStreetMaps I've found it far behind both.
What's more every company I've worked at since has paid Google for access to their Maps API.
I guess the experience made me skeptical of people proclaiming that someone does or does not have a moat because the community will just eat away at any commercial product.