aerodude

Having a quick read, they're pointing out that the Bernoulli distribution is only supported for values of 0 and 1 (i.e. it's binary), whereas pixel values for a grayscale image are a decimal value in the interval [0, 1]. When you train a VAE, it's pretty standard to use a BCE loss, but this is wrong because the data isn't binary (i.e. it's not a Bernoulli distribution). They define a continuous analogue of a Bernoulli distribution that is supported in [0, 1], and use this as the loss function for training a VAE, which gives them reconstructions that are closer to the input.

The number of subgraphs increases exponentially with the number of additional layers (and neurons). If you started off with a network the size of the final pruned network, you would have a dramatically lower chance of finding a winning ticket compared to the oversized network you start with.

I'd go even further than that and argue that the reason vegan and vegetarian diets have taken off as much as they have is because they became status symbols in the 90s / 2000s.

Rockets have an exhaust with constant velocity, and they expel mass in the form of fuel. It's this expulsion of mass that generates the force that propels the rocket. Taking a force balance and assuming no external forces (e.g. gravity), you can integrate to get the rocket equation, which tells us how much fuel mass we need to reach a given velocity (the quantity we care about when talking about stellar distances).

>If you're in the USA and fill your tank once a week, you probably spend ~$2000/yr in gas. And that's how much you'd save in gas if you went electric.

The electric bill makes a big difference, because you're really looking at the relative cost of two different forms of energy -- electrical and chemical (fuel). The last time I calculated it, the cost of running an ICE was around 20% more expensive (in my area) than a BEV, but a PHEV was within a percent or so of owning a BEV.

I'm curious to know what percentage of EV owners in Norway have a second car. I'd imagine the uptake rate for families that can afford to own more than one vehicle is significantly higher than for those that can't, but I've never seen any stats on the matter. The reality is that for most people, a car is the second largest purchase they'll ever make (if not the largest), so utility has to be high on the list of priorities. I'm not entirely convinced EVs have the utility to overtake ICE vehicles just yet, especially when incentives are removed.

I think this is the big one. Consoles and CoD dominated the market for a long time. Has the pendulum swung back towards PC games? I have to confess, it's been a long time since I played anything. Games dev companies just stopped making games for people like me.

>50km a day in charge capacity is the absolute best-case scenario. In reality you should count on half of that.

I'd count on even less than that. Industrial solar farms average a capacity factor of around 20-25%, dropping as low as 10%. And that's with farms built for maximizing sunlight. A car is likely to do far worse, especially when you consider imperfect conditions, and how frequently people park indoors. If it's a tossup between protecting my car from the elements, and getting a few miles a day of charge, I'm going to choose the former.

> So successful products optimise for the UX of a user who doesn't yet know how to use the product well. And such users really love touchscreens.

I think the problem is that touchscreens get ported to applications where there should be a reasonable expectation that the end user is an expert in the system. For example, cars, and aircraft. Touchscreens are great when you have portable systems that have to condense a lot of functionality into a small device, but I don't want to be in a position where a pilot has to touch the correct button on a touchscreen in the middle of serious turbulence. Likewise, no driver should be taking their eyes off the road to navigate to the air-conditioning tab. Applying touchscreens in these situations is not only bad engineering, it's outright dangerous. You have to demonstrate competent control of a vehicle just to operate it, so we shouldn't be assuming operators are brand new users that aren't committed to the product.

Wouldn't it be easier to point a camera at the windshield and detect a change in opacity? I would think that's a far easier problem to solve than, say, facial recognition.

Wouldn't both of those solutions undermine the advantage of the airframe, which is the fuel cost savings of the larger engines?

That said, I think seating pitch and spacing is definitely a conversation that needs to be had. I'm an average-sized guy, and I have a hard time with today's airline seats. I would hate to see what it's like for someone taller. There definitely needs to be more stringent regulation of airline seating to mitigate ever-vanishing personal space.

> Also, if you have infinite energy, there are many things you can do to cool off. Everyone will be able to set their own preferred temperature.

Jesus. Someone failed thermodynamics.

Ummm... if you have infinite power to pull carbon from the atmosphere, don't you also have infinite power to launch radioactive nuclear waste at the sun using mass-drivers? Sounds like that would be a lot more efficient than what you're proposing, too, since it's all fairly concentrated.

The whole infinite power from space argument seems problematic. Even assuming you just meant "a large, globally significant amount of power", I'm guessing that the inverse-square law and heat buildup would sink the idea. I can only imagine the dead zone surrounding the location where we beam petawatts of radiation down from orbit. It would make our worst ecological disasters look like a joke in comparison.

Faster hardware will help, but I'm not convinced that it's the answer. OpenAI Five used on the order of 2000 years of experience to train their agent. There are clearly still huge algorithmic gains to be had.

Given how we've managed to improve on nature in other domains (see solar cell efficiency, for example), I think that if we can figure out how intelligent organisms manage to learn so quickly we can likely beat nature's efficiency.

I think this is the real answer. When we developed flight, the measure wasn't "can we fly like birds?" We still haven't achieved that even today, but we fly in otherwise unimagined, but equally powerful ways.

We seem to be looking at intelligence in humans and thinking we need to develop that, without first defining what intelligence actually is. We don't exist in isolation, and it's likely that the components of intelligence exist to varying degrees in other organisms. In the same way that birds, bats, gliders and insects all have wings that generate lift, what are the things that we have in common with other animals?

For the majority of people, a car is a utility item; car buyers tend to price in the worst case scenario when buying a new vehicle. Statistical arguments don't really work in these cases, because even if an EV covers 90% of your use case, if the 10% is critical, the EV isn't an option.

As an example, a tradie who runs their own business might very well rack up enough hours driving that an EV just wouldn't work for them (not to mention the likelihood of carrying tools or towing trailers). Or what happens if you get home late and the street charger is taken?

EVs seem great on the surface, but are they necessarily practical for the average person? I mean, they could be an iPhone moment, but they could also be an Amazon delivery drone moment, too.