The specular highlights on faces definitely look wrong to me though I struggle to describe why. Shadows and diffuse lighting is a totally different story, though. Look at how it completely deletes the shadow of the steeple on the right hand side[1], or how it completely eliminates the shadows on this guy's face and jacket. Overcast lighting is an easy cheat for hyper-realism[3] and almost every single scene shown has softened or absent shadows and more diffuse light.
As an aside, I'm starting to wonder if they are modifying engine settings when switching it on and off. There's clearly some amount of accumulation it has to do and its impossible to frame-by-frame a video of a monitor, but in [1] the first frame snaps from a dynamic shadow of the steeple to a generic small blob shadow, then gets entirely eliminated on the next frame.
Hmm, I do see the shadows being removed in the links you have, and have noticed that the backgrounds do look like their lighted differently from the original, but was wondering if that is just because the AI lights things differently? - they did say that these AI effects are done with the actual 3d assets themselves and is not just some type of filter that run over the existing images, so could see how the lighting could change quite a bit.
Yeah, may be the fact that they are lighted differently from the original is turning people off. Understandable. For me, still find it impressive, and think the level detail in the faces and clothing is a full step up in capability.
> they did say that these AI effects are done with the actual 3d assets themselves and is not just some type of filter that run over the existing images
That was essentially just Jensen Huang lying during his Q&A. DLSS5 uses the same input data as DLSS<5, which is just screen space color data and motion vectors. From NVIDIAs announcement: "DLSS 5 takes a game’s color and motion vectors for each frame as input, and uses an AI model to infuse the scene with photoreal lighting and materials that are anchored to source 3D content and consistent from frame to frame."
I agree, every shot has something to like, especially in fine details, but I question the feasibility of fixing the issues while running the model on a consumer GPU in realtime. Getting similar improvements without falling back to diffuse lighting would require the model to infer a huge amount of information about off-screen light sources and objects. I'm much more excited about putting my tensor cores and vram towards neural textures since they can actually add detail at the geometry level.
Hmm, actually heard it from a podcast that this is actually working with the 3d assets, and even that statement you quoted says "anchored to source 3D content." Although, that could mean a lot of things and it's still early on, so it could still just be a pass at the end by an AI model. Yeah, I'll stay on the fence until more details are released - and should mention, am no graphics expert, and am only giving my opinion as a fan of good graphics on what the results look like:)
I think the charitable read is that Apple wants to minimize confusion by ensuring all usb ports on the device have the same capabilities. A simple USB 2.0 port would be cheap but supporting charging and thunderbolt would add meaningful cost.
edit: NVM lol, the Neo only has one fully featured USB port
For fixed route transit, speed is latency. The faster the bus can make the average trip, the tighter the timetable can be given the same number of buses. Fewer stops also improves consistency which means you can plan to arrive at the stop closer to the scheduled time, and timetables can be tightened even more by reducing the layover times that keep the bus synchronized with the time table.
Separately, the variability problem can be somewhat solved with the real-time location updates that many agencies provide. You'll still have to wait the same amount of time, but some of it can be done comfortably in your house when the bus is running late.
Sections of lines that already have meaningful congestion at adjacent stops wouldn't be a good target for balancing. WMATA in D.C. recently eliminated about 5% of bus stops as part of their overhauled bus network, this is how they described their strategy[1]: "We thought carefully about each stop, looking at things like how many people use it, how far away it is from the next stops, and whether it's safe to walk there. We also listened to feedback from thousands of bus riders."
Additionally, many stops with a lot of people loading and unloading are hubs which would never be balanced away, and often are designated timing points where the bus will wait to get back on schedule, so loading/unloading time is often irrelevant because predictability is being prioritized over speed. Improving speed and consistency with techniques like removing unnecessary stops increases predictability and allows for tightening up timetables and minimizing average hold times.
Demand-responsive transport (DRT) has been tried a bunch of times in all sorts of different environments and pretty much never lives up to the promise. Predictability is really important and ridership drops as soon as users start having to plan too far ahead, which in the past has been essential to DRT routing.
Autonomy could improve responsiveness to demand but you still run into other issues. DRT usually won't be able to take advantage of things proven to make buses faster and more consistent (bus lanes, reducing stop count, transit priority signals). Futher, consistency and response times gained by dynamic routing can easily be overshadowed by increased variability in trip time as the route adjusts to add new passengers or make out of the way drop-offs.
I've seen it work pretty well in a number of places in the form of privately owned minibuses/vans that can rapidly go where the demand is needed.
As an example, all throughout the Eastern Caribbean this system works really well (in my experience better than most centrally planned bus systems in large cities). On any given island you can go to any main road and within a few minutes a minibus will come along. Most of the time if your aren't familiar with the geography, you just tell the conductor where you are trying to get to, and they will make sure that you get off in the right spot to get where you are going or connect to another minibus. Typical cost was ~$2.
Predictability was pretty low, but because of the small size of busses, there were a lot of them roaming around, I don't think I ever waited more than 15 minutes, and that was in very out of the way places.
It's really not ideal. Similar systems are common in Central Asia. They make it difficult for travelers to predict journey times, it's unfriendly to tourists, and it's much less accessible to other populations (e.g. the disabled). They also don't scale well to large urban environments or out of the way journeys in my experience.
Yes, like all systems, it has tradeoffs. Although I would argue that some of the downsides you highlight are worse with traditional bus systems (e.g. the Caribbean bus conductors will happily guide tourists, and I have seen them go off-route frequently to drop off someone with limited mobility. Large cities in other parts of the world have managed to scale the system out to fill in gaps with other forms of transit like Lima, Peru)
The GP was arguing that it NEVER works out, and I'm just pointing out that it does work in many places.
I would much rather rely on the Caribbean minibus systems than try to rely on transit in cities like Phoenix.
To standardize a journey time in a scheduled system, you subtract the origin scheduled arrival from the destination scheduled arrival. Map apps will even do this for you automatically. If the bus is unreliable, you add error margin. A demand shuttle system usually has a much larger variance, which means you can't predict that the journey time will be acceptable and you'll find some other way to get around.
How do buses fair in this regard?
You look at the route map and the schedule to decide? Again, map apps make this trivial for regularly scheduled services.
> has been tried a bunch of times in all sorts of different environments
Has it? When, where and with what technology?
> Predictability is really important and ridership drops as soon as users start having to plan too far ahead
Uber etc have proven this to be patently false. Existing buses are experiencing dropping ridership - Uber is not.
> won't be able to take advantage of things proven to make buses faster and more consistent
You're replacing buses with auto-shuttles. Just let the shuttles use the bus lanes.
> bus lanes, reducing stop count, transit priority signals
All of these are usable if you widen the scope to include auto-shuttles.
> consistency and response times gained by dynamic routing can easily be overshadowed by increased variability
What is the difference between Busing and Shuttles here? A bus user can keep yanking the stop cord, there can be 1 or 2 disabled passengers who take several minutes to board, there can be 50 children getting on / off. These issues are constants and all are improved with demand based shuttles.
On top of what others have said, the Watchface/App dev experience is pretty great. The OS provides a lot of compositing and animation features that encourage really lively and cute designs, and the Pebble app has a JS runtime that allows apps to do whatever phone-side stuff you need without having to build separate Android and iOS apps (or, as a user, install a ton of companion apps). Spin-up and iteration is really easy because pebble-tool manages building, deploying to QEMU, and running the phone-side code in Node.js so that you can launch and test your app end-to-end with one command.
Having to write C on the watch-side isn't everyone's cup of tea but they are actively working on a replacement for rocky.js so that you can write everything in JS.
Lemonade doesn't support your claim that FSD is a safer driver than you are. It just says that, most charitably, they believe FSD and a human operator are safer than just a human operator (The co-founder said exactly this to Reuters [0]). Further, the program has only been around for a week and their marketing copy specifically cites "Tesla's data" as the source for the 50% reduction rather than any sort of independent analysis.
They are putting their money behind their words, unless there is some backroom deal we don't know about. If a human operator + FSD is twice safer than human operator alone, then FSD is still a large safety improvement. Considering how human operators behave with these systems, I'd also wager having the human operator (many don't even look at the road!) makes only a small difference.
> They are putting their money behind their words, unless there is some backroom deal we don't know about.
Their product is dynamically priced and individualized, and there is no guarantee of what the base rate will be. I don't see any reason they can't keep offering the 50% discount and then adjust the base rates to reverse engineer a sustainable price regardless of FSDs real safety.
> Considering how human operators behave with these systems, I'd also wager having the human operator (many don't even look at the road!) makes only a small difference.
Lemonade will likely be getting driver monitoring telemetry and calculating rates accordingly, but in either case I'm convinced that we are still on the left hand side of the Valley of Degraded Supervision [0]. Operators may not pay full attention at all times but they likely still have pretty good heuristics for what situations are difficult for FSD and adjust their monitoring behavior accordingly.
Tesla could of course release detailed crash and disengagement data to prove FSD safety. That they do not is itself a form of evidence, and in lieu of that we have to rely on crowdsourced data which says FSD 14.x still has a very long way to go to be safer than the average driver [1].
It is in the EU but in the US ADAS won't be mandated until 2029. It would tank your IIHS rating though and all major mfgs have met a voluntary pledge to have >95% light duty vehicles ship with autobraking by 2023: https://www.iihs.org/news/detail/automakers-fulfill-autobrak...
Any time you are reading a law, especially one from another jurisdiction, you have to be very careful to consider that there may be terms with a legal or common law definition that you don't understand. In this case, "reckless" seems to be a well defined term with a fair amount of case law behind it. To my untrained eye it seems like a newspaper would be well within their rights to publish harmful information as long as they avoid "a conscious disregard of a substantial and unjustifiable risk".
As an aside, I'm starting to wonder if they are modifying engine settings when switching it on and off. There's clearly some amount of accumulation it has to do and its impossible to frame-by-frame a video of a monitor, but in [1] the first frame snaps from a dynamic shadow of the steeple to a generic small blob shadow, then gets entirely eliminated on the next frame.
[1] https://youtu.be/4ZlwTtgbgVA?t=435, [2] https://youtu.be/4ZlwTtgbgVA?t=326, [3] Cyberpunk hyper-realism mod: https://www.youtube.com/watch?v=_toA8lErAHg
reply