Numlock News: April 5, 2021 • Paleobotany, Solar, Godzilla
By Walt Hickey
Godzilla vs. Kong made $48.5 million in its first five days over the course of the Easter weekend domestically, the best performance of any film released over the course of the pandemic. That’s vastly higher than the $16.7 million brought in by Wonder Woman 1984 over Christmas, which is somewhat understandable given that now over 106 million Americans have received at least one vaccine dose. The film is a classic buddy comedy, putting the prickly Kong and free-spirited Godzilla together in a quest to renovate downtown Hong Kong.
The Current Box Is Full
The pandemic has prompted a boom in sales for Pokémon cards, and the authentication service that vouches for the quality of various cards — Professional Sports Authenticator — has announced they’re straight-up pausing new business in order to deal with a colossal backlog. According to PSA, over 1 million Pokémon cards were submitted for grading in 2020, and in 2021 they are already on track to quadruple that. PSA said they received more cards in three days of March than they had in the previous three months amid sky-high sales juiced presumably by voracious coverage in numbers-driven daily morning newsletters. Pokémon cards now account for 30 percent of card submissions to PSA, up from 13 percent in 2020. They will accept some cards, but they have to be worth more than $2,500.
Netflix is heavily invested in the voice dubbing business, having lined up 170 dubbing studio partnerships to produce dialogue for translated programming in at least 34 languages. One recent example — Lupin, a French crime show that’s become the most-watched debut of a non-English language series in large part to ample dubbing — is in the top 10 list of not just the U.S., but also Austria, Croatia, Lebanon, Nigeria and Uruguay, and 86 percent of those who watched the show used subtitles or dubs. According to Netflix, the average U.S. viewer watches three times as much dubbed content on the service as they did in 2018, a dizzying loss for my fellow combatants on the subs side of the mid-2000s subs v. dubs internet war.
California was among the first places in the world to adapt net metering, which allowed people who installed solar panels on their roofs to actually sell excess power back to the grid. Since then, more people have adopted solar, which has led to increased electricity costs to those who don’t have the panels, and the California Public Utilities Commission is weighing an overhaul of the policy that could have nationwide implications. Non-solar customers are paying up to $230 more annually for electricity than those with panels, which, again, was kind of a key selling point of the program, an upfront green investment gets you savings down the line. Right now, over 9 percent of residential customers have rooftop solar, north of the approximate line of 5 percent when other ratepayers begin to notice a change in their billing. People with solar tend to be richer, and get an average credit of 25 cents per kilowatt-hour when selling to the utility, compared to the 3 cents per kWh the utility pays when buying solar energy from a large-scale farm.
Institutions with easy access to capital are elbowing individual homebuyers out of the market, buying up limited housing stock in major cities to serve as rentals. The effect? A housing market where everything is more expensive for first-time buyers. About one out of every five homes sold in the U.S. is bought by someone who never moves in. In Houston, investors account for 24 percent of home purchases.
A new analysis of 50,000 fossil pollen grains and 6,000 fossilized leaves from around 53 sites across Colombia found that after the asteroid that wiped out the dinosaurs struck the planet, 45 percent of South America’s tropical plant species disappeared. Over the next 6 million years, forests gradually grew back to the level of botanical diversity notched right before the asteroid, with the plants that endured making up a significantly different ecosystem than the one that came before, laying the groundwork for the Amazon rainforest of today.
A new study analyzed the accuracy of labeling on the 10 most-cited training data sets used to fuel machine learning models, and even setting aside many of the other myriad issues and biases that can be found in datasets fueling AI, the researchers found high levels of plain incorrect data in the sets. These are errors in labeling, such as an image of a cat being labeled “dog,” or a flower being labeled “pencil,” or an image of Godzilla or Kong being labeled “interesting protagonist.” The ImageNet test set had an estimated label error rate of 5.8 percent, and the test set for QuickDraw had an estimated error rate of 10.1 percent, with five of the other models having an estimated error rate of over 2 percent.
The best way to reach new readers is word of mouth. If you click THIS LINK in your inbox, it’ll create an easy-to-send pre-written email you can just fire off to some friends.