alexmlamb

I've used the McDonalds one in Canada and it seemed pretty good. I think the "options" button is partially an upsell, since they want to try to advertise extra stuff.

Has anyone done an analysis to see what fraction of the wealth generated by silicon valley ends up in the hands of people who happened to own land there? If normal houses cost millions of dollars, I could see it being a sizable fraction.

I think the pricing mechanism plays a useful role in encouraging people with valuable land to give it to others who will use it more productively. At the same time, I think there's a risk that people who just own land could capture an unreasonable fraction of the generated wealth.

Would it really be right for San Francisco homeowners to capture over 50% of the wealth from Silicon Valley's success?

Fake news concern isn't the right place to focus. News type content is low-volume, has a structured sharing model (i.e. people usually share the articles), is long-form, and requires fairly high precision. This is where humans excel and where automated language models are the weakest.

On the other hand, comments on sites like Facebook/Reddit/4chan are perfect for language model bots. The content is high-volume, semi-anonymous, usually shared without an explicit network, and can be extremely low precision.

So if you had a bot that could get on any discussion network for planning protests for example, and spammed it with destructive and divisive fake comments, it could actually make organizing pretty hard. And while there's some demand on the language model, many comments are just a few sentences long. And the content doesn't need to be very precise, it just needs to be convincing enough to be distracting.

I also think that the worst abusers are likely to be current power-players like Google/Facebook/Chinese Gov rather than small actors.

Many of these fall under the pattern of comedic overexpression or tactical underexpression. For example, "You're really dirty, you literally smell like trash" obviously means "you figuratively smell like trash" but the latter doesn't pack the same punch.

At the same time something like "not the best programmer" is a minimizing statement which is less mean than something like "bad programmer".

Part of the purpose of activism is to exploit current laws for immediate change, but the other part is to create support for changing the laws to provide better protections.

The idea that you could disappear completely from anyone using modern technology (which is what getting accounts banned from communication services will amount to) with no legal recourse or transparency is horrifying, even if it's technically legal, because the laws were written at a time when the tools of communication were widely distributed and could only be feasibly disrupted by governments.

In the short term, the lawyers matter. In the long term, we need to create laws and conventions which ensure that the lawyers of the future are focusing on the right issues.

I've been using Bing primarily and it seems fine. It's stronger in a few areas like image search, customization, and weather. Translation is roughly equal. For code and paper search it's somewhat worse than google but not WAY worse. I'd highly recommend making it your default search engine if only to help create a more multipolar internet power system.

It's having to write the parenthesis after print. No one wants that and it's been by far the largest complaint the community has about python3.

Yeah sometimes this is called a "subgradient".

It's interesting that some of his intuitions seem really really good like the idea that you'll get better results by using very large models. On the other hand, specialized hardware before the software works well has always been a disaster because it's so hard to iterate on new hardware. His idea of using evolution as a primary learning process also doesn't work well.

The idea of Six Sigma never made sense to me. I read a book on it, and it basically said that you should optimize a business so that mistakes are extremely rare (the six sigma refers to that point on a N(0,1) gaussian cdf).

The idea that you should try to avoid making mistakes seems like a good one, but also is rather obvious, and the right threshold is also obviously problem dependent.

I suppose the idea is kind of fluff, but of all the fluffy ideas out there, trying to be reliable and dependable is probably one of the better ones.

It seems weird to me to conflate "advocating the legal right to own technology that can be used to do X" with "doing X", even though obviously you can disagree with both.

"ridiculous situation in Hong Kong"

How is it ridiculous? I can't think of any government, except maybe Canada, that would just allow a part of their country to secede, especially if the movement to do so was explicitly backed and funded by a foreign power (in the case of HK, the same foreign power that used to occupy that part of the country).

One idea: try using adversarial training (i.e. train the classifier on examples with bounded perturbations which maximize error).

I think this is a neat idea. Basically, you'd get more "credit" if you're cited by a good paper than if you're cited by a bad paper.

I don't think it's necessarily like that. The error could make the theorem wrong in all cases.

I think it's more that the mathematician has a top-down way of reasoning, where they can see things like "I want to get from New York City to Los Angeles, so I have to board the bus, take a flight, and then take the bus from the airport at LA". There are certain parts where you basically know that a proof will be possible, because it seems true, like "I can get to LA's airport with public transit", so usually a specific hiccup, like a bus being delayed, won't prevent you from getting there.

I think a lot of people see it as a slippery slope. Of course FTP is not so common, but what if at some point they remove the ability to visit general URLs and just let you either google search or go to a page you've already visited.

Sadly a decent fraction of users might just accept something like this, even though it would be a major step backwards for internet freedom.

I'm just making a point about the abstract argument. I have no idea about whether Homebrew is technically impressive or not, but it sounds like people think that it is?

Is it just me or does this reasoning not make any sense?

"NASA: 90% of our rocket scientists use the A/C you just fixed, but you can't explain how a rocket works so f* off"

Presumably what matters is whether homebrew is technically significant or relevant, not just that they use it?

I guess you don't necessarily need data, but if you don't have data than you'd need a simulator, and we'd don't have simulators for most problems that we care about.

"Studies have shown that journal editors prefer reviewers of the same gender, that women are underrepresented in the peer review process, and that reviewers tend to be influenced by demographic factors like the author’s gender or institutional affiliation. "

Most venues are double blind, so reviewers wouldn't know about the latter things. Having fewer women in the reviewer pool is unfortunate, but it's hard to see how it would effect the mean review quality in a direct way assuming that men and women are equally good at reviewing (maybe in an indirect way it could harm reviews by increasing the load per-reviewer).

I lived in Baltimore for a few years in college. There are definitely some nice things about it, like the aquarium, and a decent choice of restaurants (however this should come with the caveat that the inner harbor has lots of police standing around, so its level of safety is not that organic). It also has a good selection of sports teams for a city its size.

https://en.wikipedia.org/wiki/List_of_United_States_cities_b...

It still has one of the highest crime rates in the US, so I don't think criticism of the city is necessarily ill-informed.

(I work at Lyft)

I've had it with you Scale.AI people always trying to take credit for Lyft's work. We've been working weekends and nights for years, even the hourly workers have had to take unpaid overtime. All that time, I've never seen Scale.AI do extra work to help us before a big deadline.

Why don't mainstream newspapers (like NYTimes, Economist, etc) include a list of citations at the end of articles the way that scientific papers do? It seems like this would be easy to do, and would help to restore a lot of trust, because people could easily see if they like the sources being referenced and if the sources are actually fairly represented in the article.

It's not exactly a quick fix, but how many generations would it take to breed cats with substantially softer claws? This could be an interesting program perhaps for the government to undertake.

I also wonder if crispr or embryo selection could be used if a GWAS could find genes that lead to softer claws.

The selection techniques have been used to improve milk yields for cows, so it seems like making cats which don't need to be declawed would be a valid project.

Just to be clear, this doesn't necessarily mean that the students got the degree through fraud or low standards.

Speaking and listening can be harder than reading and writing, so the students might be doing good work even if their spoken language isn't very good. Also in technical subjects you might not need to know the language well to pass the tests and do the homeworks.

As an example, I have an Msc from a French language university, but I can't really understand the language.

It seems like this can easily be fixed by having people sign up for contracts that last more than 1 year or at least have some financial incentives that accrue to employees who stay longer? For example, tech companies have stocks that only vest after 4 years. I don't see why in principle this couldn't be even longer.

I don't know much about it personally, but I think the US government has a lot of benefits (like retirement) that scales with years of service.

I actually suspect that at least for many big companies the problem isn't getting people to stay after training, it's getting the best employees to stay and getting lower-performing employees to leave. I think for this reason, most tech companies structure pay with a 4-year contract, which promises a high salary for 4-years (due to bonuses and stock), and then an implicit pay drop if there's no promotion after 4-years.

Well, I think that the hype is one-directional because the default assumption is that machines will lose to humans in tasks that involve reasoning. We know that there's no computer that can learn to put away the dishes after watching a person do it, so there's nothing to talk about and nothing to hype.

I don't think that this is quite the right explanation.

|y - wx| isn't differentiable but it is subdifferentiable. So the derivative is defined at every point except where y = wx, and in that case you're more or less fine if you just pretend that the derivative is zero at that point.

One reason why squaring is preferred is that the derivative has a nice closed form, which can be used to find closed form solutions.

Another is that people like to fit their models with gradient descent, and gradient descent has better guarantees for strongly convex loss functions. It also works better in practice. Intuitively: if you try to minimize x^2 by gradient descent, you have the largest gradients when you're far from the minimum. If you try to minimize |x|, then your gradient is always either +1 or -1, making it harder to converge around the minimum. See the Huber loss for a strongly convex relaxation of the absolute value function.