MSFS 2020's Earth is unbelievably good from the air, but it (understandably) does not look great and is sometimes not entirely physically coherent up close at the surface (outside of airports and a limited set of bespoke areas).

There's also some crazy errors that are visible from the air, like many rivers being raised up hundreds of feet[0], and many bridges being solid underneath[1].

I always assumed one of primary things in a sequel would be to train an even bigger and better satellite maps -> 3D scenery model, especially one that understood a wider (and better-localized) variety of building configurations, so we'll see.

[0] https://forums.flightsimulator.com/t/river-altitude-raised-w...

[1] https://forums.flightsimulator.com/t/fix-photogrammetry-brid...

"earth copy" is just marketing lingo. Models are approximations made for a specific goal at a time. You cannot build a model that is both good for gaming and meteorology.

99% of that engine value is procedural terrain and graphical shaders. It looks great but has no scientifical value.

Just like the "digital twin" hype train, cool simulations have fancy graphics, useful simulations don't.

>You cannot build a model that is both good for gaming and meteorology.

This was the prevailing wisdom in AI before generalized transformers as well. We're rapidly moving toward black box hyperintelligent AGI.

I was comparing expert systems here, not emergent ones

There has been zero motion towards true AGI so far, hyperintelligent or otherwise.

So true! I think a lot of people, even here on HN, get confused by the marketing term, "AI".

Authentic Ignorance, I guess…

I would argue that humanity has been advancing rather steadily towards "true AGI" since at least the Jacquard loom. Otherwise, if you wait to admit progress until you actually have evidence for having achieved AGI, it'll just look like the Heaviside step function.

How can you claim that we've been advancing steadily when we don't even know how far the destination is, or if we're moving in the correct direction? There's no basis to claim that it will look anything like a step function; if we do achieve true AGI someday the first one might be equivalent to a really stupid person and then subsequent iterations will gradually improve from there. It seems like a lot of assumptions are being made.

You're really not clear if we're moving in the correct direction? What then would be for you an indication that we are?

Show me a computer that can reason in a generalized way at least as well as a chimpanzee (or whatever). And no, LLMs don't count. They are not generalized in any meaningful way.

Please do clarify why LLMs aren't generalized - other than not being embodied, they seem quite general to me. Is there any particular reasoning task that you have in mind that chimpanzees are good at, but LLMs are incapable of?

That's the goal of Nvidia's "Earth 2" plans, which the article mentions very briefly.

It's Neil Stephenson's snowcrash come to life :-)

We used to be in this space, and with Ayvri/Doarama - which was a 3D virtual world but not a flight simulator, it was used by paragliders to replay flights, commercial drone operators, some defense, transportation planning (trains, maybe others as well), etc etc.

We approached Microsoft when we were selling the company, and the response we got was "you're too close to work we're doing for us to even look at you or talk to you". I assumed that meant they are building out a digital twin, with capabilities to run in a browser at high speed with high resolution (which we did). That was 3 years ago. I'm surprised we haven't seen something from them yet.

It is Microsoft after all. They are quick to put out press releases and imply or even promise functionality, but you can spend your entire life waiting for them to actually deliver.

Imagine GTA on that map. Wondering how many years away are we from this.

Mostly military simulations and training, probably

The lighting, clouds, weather are great and make it look beautiful. Nature looks great in the game. But in terms of urban areas, particularly buildings and roads it is much much worse than Google Maps/Earth. You also see the seams in the construct in places like coastlines.

That used to exist as a product called Microsoft ESP. It got sold off to Lockheed Martin when the Flight Simulator project was shut down.

https://en.wikipedia.org/wiki/Microsoft_Flight_Simulator#Loc...

I would love a real world driving sim. I don’t really care for driving on race tracks, but real roads.

I've always said, a version of Grand Theft Auto, but with a real map of [your city], and with cops that enforce driving rules, would be a priceless tool for teaching young drivers.

After I passed driver's ed, I still spent the next two years getting lost within a few miles of my house. But drop me in any alley in GTA5 and I can get to any other spot in a pretty short path (although I confess it often does involve dangerous jumps).

Midtown Madness 1 & 2 are most similar to what you are looking for.

it's one of those things I'd do as a millionaire.

Some people would buy boats, private planes, art, ... I would hire a team to replicate my child neighbourhood perfectly in a gran turismo or preferably GTA 5 engine lol. You'd get pretty far with a million USD in the modding community

Lol, if I were a millionaire I'd probably buy half a house somewhere kind of nice.

You’d get flewced pretty quickly in any modding community.

They could live stream the local radio for the soundtrack.

I made a basic version of this to help me pass the UK driving test, which is fairly difficult: 40 minutes of virtually error-free driving is required and the pass rate is 50%.

It's 2.5D, I used a 3D engine but the graphics are just satellite imagery projected on to a terrain map.

>> would be a priceless tool for teaching young drivers.

And an amazing resource for organizing street racing or any other event where knowledge of average police response times might be useful. Playtesting exactly how one can best evade pursuing police on realworld streets would be very fun.

American Truck simulator comes close. But something like that bootstrapped of open street maps could be awesome.

American Truck Simulator has 300 cities and it takes 3 hours to drive across the map in real time. It's probably the biggest world in a driving game.

The cities (at least for ETS) are a miniature version with only a couple of streets and some landmarks (in larger cities). It doesn't really come close to reality.

BeamNG comes remarkably close in experience, although not in real world. I would love to have MSFS world as a driving game like BeamNG.

Check out Euro Truck Simulator then, I hear it's very good.