It's possible they want to position it as something other than a search engine, like "answer engine" or "personal agent" or whatever, even if what it's meant to do is replace the traditional search engine.
In the city where I live, also in the Netherlands, we had multiple of these but all of them got converted back to normal intersections with traffic lights and it has been a huge improvement.
My understaning is that normal intersection with traffic lights has higher troughput (as cars can just fly through it at or near 50 km/h if the green light is long enough), but roundabout have lower average wait times (if the traffic is low to moderate, you usually wait less that 5-10 seconds to enter the intersection).
No, simple roundabouts are worse in both throughput and wait times than normal intersections with traffic lights. They are however safer, and that safety is precisely thanks to slowing traffic town, which means that collisions are less frequent because drivers have more time to react and if a collision happens it happens at lower speed. Just think about it: how can wait time be smaller and throughput higher when cars approaching a roundabout from _any_ direction need to slow down compared to normal intersection where cars on one road can wizz through with full speed?
To answer your question: wait times can be lower because, if the traffic is low enough, ALL cars just wizz through, at the cost of slowing down while entering the roundabout. Whereas, for the intersection with traffic lights, a significant number of cars will hit red light and have to wait for a minute or two.
Except it's not true. You can't wizz through a roundabout because the road is curved, you are physically unable to drive through it with the same speed as you drive on a straight road. It's by design. When traffic is very light then the traffic lights can be simply disabled (for instance, at night). Not disabling them is the failure of the people responsible for the traffic lights, not some inherent unsolvable problem.
This depends on how light the traffic is. In a lot of places, traffic outside of rush hours is light to moderate (but definitely not light enough to justify disabling the lights), and the roundabouts might the more efficient solution, at least outside of rush hours. It was probably studied by someone already, as this would make a decent research topic for any transporation scientist.
In light traffic roundabouts are actually worse than normal intersection. How can it even be disputed? Let's say you have one car per hour, or any number of cars approaching in such a way that none of them gets close to the intersection at the same time. It goes through normal intersection without much slow down. It goes through equivalent roundabout _always_ slowing substantially down. What's here to dispute? If you have cars approaching at the same time then it gets a bit more complicated but the claim that in such a situation you can still always beat normal intersection is just ridiculous, because if the cars show up at exactly the right time some of them still need to completely stop before entering roundabout because there is already another car circling it. Analogously with normal intersection you don't need to always stop before entering it because both cars might be turning into perpendicular roads and they don't cross their paths at all so they both can go smoothly through without stopping at all. I do agree that roundabouts make the intersections safer, that's great, but it's at the cost of throughput but everybody pretends it's not true. Even the claim that the delay is reduced because cars would need to wait on traffic lights is dubious, because you can't just enter or leave a roundabout when there is heavy traffic through it that competes with your 'route' , roundabout is not equal for all entry points and doesn't distribute all of them equally. So you might wait just as long as you would waiting on traffic lights. Standard intersections are worse only if they have dumb traffic lights and the traffic is light, that forces you to stop and wait even though there is no competing traffic. But again, it's a case of fixing the lights and making them sense the incoming traffic instead of installing roundabouts.
> Let's say you have one car per hour, or any number of cars approaching in such a way that none of them gets close to the intersection at the same time.
That's not light traffic, that's basically no traffic. I don't think anyone bothers to build a roundabout for such intersection.
Roundabouts are for light to moderate traffic scenarios. Their main benefit is that they get drivers to slow down, which vastly reduces number of accidents and fatalities on the intersection. The effect on throughput is secondary, as the intersection was not congested anyway.
That claim of reduced fatalities is also getting outdated now, since the introduction of roundabouts with more than one line. Also accident rate for the cyclists is actually higher on roundabout than in a a standard intersection. So all in all, roundabouts are not as great as many insist and my beef with them and their supporters is that they keep bringing up the same talking points that were presented _before_ roundabouts started to be built on a massive scale and don't want to hear the empirical evidence about how these roundabouts actually work in practice.
I drove through some when visiting Aruba (Netherlands overseas province) and loved them. They seemed even safer than typical roundabouts and super intuitive
The paper was definitely cool but doesn't allow you to run particularly large LLMs on iPhones. It allows you to run a certain kind of LLM (sparse ReLU based LLMs) whose weights are somewhere less than 2x RAM. So, 7b Falcon works, but the competitive-with-gpt-3.5-turbo LLMs are still out of reach (and aren't ReLU based, although maybe that could change in the future). And nothing is competitive with GPT-4 right now.
Of course in the long run I think it will happen — smaller and more efficient models are getting better regularly, and Apple can also just ship their new iPhones with larger amounts of RAM. But I'd be very surprised if there was GPT-4 level intelligence running locally on an iPhone within the next couple years — that sized model is so big right now even with significant memory optimizations, and I think distilling it down to iPhone size would be very hard even if you had access to the weights (and Apple doesn't). More likely there will be small models that run locally, but that fall back to large models running on servers somewhere for complex tasks, at least for the next couple years.
> And nothing is competitive with GPT-4 right now.
You mean nothing available? Or you mean nothing that public knows exists? The answers to those two questions are different. There are definitely products that aren't available but the public knows exist and are upcoming that are in GPT-4's ballpark.
I mean nothing that is able to be benchmarked and validated by third parties is GPT-4 quality. I know there are upcoming releases that are hyped as being equal to GPT-4, e.g. Gemini Ultra, which I am very excited to get my hands on — but regardless, Ultra is not small enough to run on phones, even using the sparse ReLU flash memory optimization. And we'll see how it benchmarks once it's released; according to some benchmarks Gemini Pro has somewhat underperformed GPT-3.5-Turbo [1], despite Google's initial claims. (Although there are criticisms of that benchmarking, and it does beat the current 1106 version of GPT-3.5-Turbo on the Chatbot Arena leaderboard [2], although it slightly underperforms the previous 0613 version.)
I heard rumours of these claims a few weeks ago, I assume they are talking about the same thing. Nothing concrete but from a reputable person and honestly with how well mixtral performs on the chatbot arena elo board I wouldn't be surprised if it's true.
reply