cf

I think there is a good argument to be made for crowdsourcing some of those as well.

I will second this as someone who really likes these two pens and was disappointed in the Preppy.

Not to be too snarky but to the question in the article, “The question for health systems: how much are one-time therapies worth?”, you mean like vaccines? Presumably, it should cost a similar amount.

Would anyone be able to recommend the opposite of this in terms of a wakeup light? The problem with lots of these bulbs is they aren't very bright and these days I live in a place that doesn't really get lots of natural light.

Let me try to give a high-level explanation. As we have had success with using neural networks for different tasks, people began to ask what are the properties that make neural networks work so well.

When I mean work well, I mean a family of models which are generally easy to describe as well as be generally easy to train. Nearly all neural network models fit this bill as a they can be described with a loss function. A loss function being something which scores how accurately the current NN performs with a particular set of parameters. This loss function is then usually differentiable, and thanks to autodifferentiation tools, you can obtain the derivative of this function with respect to some parameters in this function. With this derivative, you can go ahead and run SGD to find the optimal parameters. We have now both empirical and theoretical evidence that SGD works fairly well for optimizing functions, particularly the high-dimensional ones neural networks describe.

So, the insight is we need a name to describe all these models which are differentiable and nice to train because of that. So now we can say what really ties CNNs, MLPs, RNNs, etc is not some biological metaphor, but that these are all expressible as some loss function we can find the minimum of using SGD.

Is there anything like Electron from a cross-platform and ease of development standpoint but way more lightweight? I often find applications written in it to be very sluggish and resource intensive even when the web or mobile version of the same application is significantly smaller and faster.

Interestingly, Kademlia ended up being a major component in lots of botnets for a time like https://en.m.wikipedia.org/wiki/Storm_botnet

In practice, SomeCoolNewTool will be bundled with some seed hosts.

Incidentally there is a workshop this Sunday at ICML about program synthesis https://uclmr.github.io/nampi/

Thanks!

This reminds me a lot of https://via.hypothes.is/ . Could someone familiar with both explain why to prefer one service over another?

Totally, but there are unlikely to be many ways to structure the law to give users consent over their data and not burden business as usual. Even responding to DMCA requests is burdensome. You can automate it like Google has tried but it fails sometimes and has become a major technical and personell challenge.

I don't want to trivialize compliance. Even ostensibly simple requirements are never quite that, and every second spent on them is time not spent on your product.

I don't get the hand wringing. If you have an image sharing service you will have to contend with child pornography. If you have any content sharing service you have to deal with copyright infringement. You need a lawyer to create a company. If you do open source you will often put a license file in your github repo.

Tech has already and will continue to interact with laws/lawyers. At some point open source libraries will appear to streamline compliance. For now it sucks but ya gotta muddle through or call it a day.

You have to be strategic if your country doesn't care, but the first step is to create the intellectual edifice activists in the future will lean against. Also even as someone not in your country, I would still enjoy reading any writing on this.

Lots of people care about this. I think the solution is to produce policy papers about this risk complete with proposals for how to go about curtailing those kinds of abuses. The EU still cares a lot about privacy, and other government bodies are happy to help if you can suggest how.

Exactly. I think the Algorithm Design Manual does more to directly train that kind of thinking than many books.

Now if you want theory, I continue to recommend Algorithm Design by Kleinberg and Tardos.

I love it since it has actual C code for implementing the ideas. You get the intuition for the algorithms, how that intuition translates into pseudocode, and how that pseudocode becomes real C code.

Another lesson that it uniquely provides is that in day to day life you don't use these algorithms. You create new algorithms using the ideas behind the famous ones as a guide. An algorithms book or class that doesn't teach you that problem-solving mindset is basically useless.

I re-implemented lots of basic unix tools. What I would do is go for a command, and just look through OS api references seeing if I could implement it. Once I got basic functionality, I would go look into the source code and check I did against the standard. I found it super-helpful.

There are ideas that only get expressed in person. Either because they are still a work in progress or because they are not interesting enough to warrant a video or a blog post. Conferences are really great for learning about that sort of stuff.

Additionally, most online communication is intrinsically asynchronous. It's hard to have a chat online that's anywhere as fluid as what happens in person. Even the flurry of twitter replies doesn't quite compare.

Many celebrities and entertainers can bring their fanbase with them as they get older. This is literally what people like Leno and the Rolling Stones do.

As someone who commutes to work on a bike, you definitely have a sympathetic ear with me.

Inventing a better battery is something a single person can do in the lab. I worry that you are conflating inventing things and the engineering of scaling them out. Scaling things almost never was doable by one person, but the initial seed was and usually is.

Inventions are motivated by problems, and we still have lots of them. We sorely need cheap sustainable energy, cheap drugs, artificial organs, more efficient transportation, denser batteries, and that's just stuff off the top of my head. Some of that stuff will entail using existing technology, buy much of it will require new inventions.

Many ideas that seem easy to invent only seem so in retrospect. The article repeatedly points out that inventing rope isn't obvious.

Buying lots of nice things on a credit card and racking up debt is a wonderful short-term strategy. At some point the mistakes you make in the short-term cripple you in the long-term. That's what all these efforts are trying to address. How do we make things that are costly long-term costly in the short-term as well?

I think this is a good application paper for Lean showing how you can use it to develop machine learning algorithms.

https://arxiv.org/abs/1706.08605

Sure, I agree with that. It's not clear the comment I'm replying to does so. If we want to view the market as a phenomenon like soil erosion that's fine. I got the impression that the parent comment was implying markets as a provably optimal solver for resource allocation problems.

See arbitrages, asymmetric information, collusion, and price fixing

I'm not sure I agree with some of these figures. Vancouver and Nashville are somehow considered to have similar cost of living.

https://www.numbeo.com/cost-of-living/compare_cities.jsp?cou...

What are people doing to cycle through windows on nested tmuxes? I often hide the outer tmux captures keybindings before the inner one can see them.

The keyword you are looking for is AutoML. http://automl.info/