turingbike

people want to go back to monoliths, but don't want to lose the upsides of microservices. once you get a taste of serverless, it's really hard to go back to overprovisioning. at the same time, distributed state and logic that crosses service boundaries is hard. I don't know what comes next, but what I _want_ to come next is powerful static analysis that works across languages and services, that can "compile down" to serveless services

I bounced a bunch of times too for similar reasons.

Is it worth checking out? emphatic yes

Summarize? I will try, and probably not do a great job; I'm still digesting it.

The world is both "nebulous" [1] and patterned; observed patterns aren't really out there so much as a combination of "there" and in our mind. Some of these patterns are incredibly useful. One pattern is "the rational / modern world view", which lets science work. It is very good, but not complete.

Post-modernism noticed that no pattern is a perfect fit on the partially-nebulous world, and went nowhere with that, and ended up in a nihilistic, bad place. That isn't good.

The goal is to see patterns as useful, conventional truths, not ultimate truths. Then you can pick the right pattern / conceptual framework with which to approach a given situation.

[1] https://meaningness.com/nebulosity

The Fed's policies (at least since 2008) could be described as "socialism for the management upper class" - they got free money that they gave to themselves as bonuses. Their companies were buoyed up, regardless of what the market wanted or whether the business was sound.

There was a great article on the front page yesterday, When to Assume Neural Networks Can Solve a Problem, https://news.ycombinator.com/item?id=22717367 . Case 2 is very relevant here, basically: if you have solved the problem with access to lots of data, usually you can adapt to a lower-data regime.

I am way out of my element talking about brain surgery and sensors. However, one thing that I do well is say "you shouldn't bet against neural networks", which is a great way to be right on a few-year time horizon.

Given this perspective, I would postpone investment as long as possible then.

I think you will also like this quote/thought then:

The theory is this: Infinite Jest is Wallace's attempt to both manifest and dramatize a revolutionary fiction style that he called for in his essay "E Unibus Pluram: Television and U.S. Fiction." The style is one in which a new sincerity will overturn the ironic detachment that hollowed out contemporary fiction towards the end of the 20th century. Wallace was trying to write an antidote to the cynicism that had pervaded and saddened so much of American culture in his lifetime. He was trying to create an entertainment that would get us talking again.

- http://fictionadvocate.com/2012/09/19/the-infinite-jest-live...

I assume you have never explored your own literal blind spot [0] and how the brain just plasters over it with some hacky texture filling...

[0] https://en.wikipedia.org/wiki/Blind_spot_(vision)

Google's AutoML produces black box models that are only available over a network call. This services seems to produces downloadable models, and a notebook with Python code that creates the model. If that is the case, this is substantially better than GCP's offering.

AWS consistently releases similar products after GCP... but they are much more well-thought-out, as AWS has to support them indefinitely...

This is called regulatory capture. Once you know the term, you can find some good content on it, like this https://www.thisamericanlife.org/536/the-secret-recordings-o...

They give examples of problems their model could solve that Mathematica couldn't (within a 30 second timeout) - and that's awesome. Destroy Mathematica. But, I did anyone notice if there were problems that it couldn't solve that Mathematica could?

I was hoping that this would be related to Colah's "Neural Networks, Types, and Functional Programming" https://colah.github.io/posts/2015-09-NN-Types-FP/ (but it is not). In it, he establishes correspondances between different NN architectures and different FP tricks like `Zipped Left & Right Accumulating Map = Bidirectional RNN`.

Does anyone know any other resources that take this approach?

I think the reason the parent mentions GraalVM instead of Jython is that Jython can't use C libs like numpy. But GraalVM lets you compile JVM to a native binary, so maybe there is some way to use Cython in JVM.

* Search "Who's Hiring?" for remote - https://hnjobs.emilburzo.com/#%22remote%22

https://weworkremotely.com

https://remoteok.io/

and of course, word-of-mouth / loose social connections.

Leo Breiman [0] (inventor of bagging and random forests) wrote a paper called "Statistical Modeling: The Two Cultures" [1], and since I read it, I see it everywhere. The basic idea is that Statisticians place(d) too high an emphasis on model interpretability ("data modeling" in the paper), and as a result, missed out on the revolution of machine learning ("algorithmic modeling" in the paper). In the author's words (parenthetical added by me), "[T]he focus in the statistical community on data models (simple, interpretable models) has [l]ed to irrelevant theory and questionable scientific conclusions."

In this TDS post, the author says "Statisticians and Actuaries are at the bottom of the heap as a prior role for existing data scientists." Maybe this isn't a coincidence? Plenty of companies had statisticians on staff, but the explosion of data science happened anyway. Why? Because data scientists do the same types of tasks as statisticians, but while statisticians are of the data modeling culture, data scientists are expected to be of the algorithmic modeling culture. It seems that the market is saying that the algorithmic modeling culture is getting results.

The author references "Type A vs Type B Data scientists" [2], which seems to be getting at the same thing: "The Type A Data Scientist is very similar to a statistician... Type B Data Scientists share some statistical background with Type A, but they are also very strong coders and may be trained software engineers. The Type B Data Scientist is mainly interested in using data "in production." They build models which interact with users, often serving recommendations (products, people you may know, ads, movies, search results)." For whatever reason, there is a correlation between Algorithmic modeling / Type B and "getting things done".

[0] https://en.wikipedia.org/wiki/Leo_Breiman

[1] https://projecteuclid.org/download/pdf_1/euclid.ss/100921372...

[2] https://www.quora.com/What-is-data-science/answer/Michael-Ho...

This is a good first-order approximation, but there are other cannabinoids which influence the effects: CBN, THCV, and others [0]. Additionally, based on first-hand anecdata, I think there is good reason to believe that the terpene profile can have a big effect on the experience, as well as the flavor [1].

But, yeah, if THC causes anxiety for you, try to never go above a 1:1 THC:CBD ratio. For comparison, most sativas are around 200:1 (20% THC : .1% CBD). Since CBD is legal over the counter, you can usually enforce this ratio, regardless of the THC source.

[0] https://www.leafly.com/news/cannabis-101/list-major-cannabin... [1] https://www.leafly.com/news/cannabis-101/terpenes-the-flavor...

Stamets' book, Mycelium Running: How Mushrooms Can Help Save the World [0] is very good, and his enthusiasm is definitely contagious. I will check out the podcast, thought tbh Joe Rogan is pretty hit or miss for me.

[0] https://www.amazon.com/Mycelium-Running-Mushrooms-Help-World...

Named parameters _might_ be my favorite part of Python. I don't understand why every language doesn't have them.

This is the funniest satire that has ever made it to the fp of this site. I was literally LOLing at the videos.

I have been looking for exactly this content for years! Thank you!

What if future humans check all the planets in the universe

Nope! Imagine a shell around us, defined as "how far we could go, if we left right now, traveling at near the speed of light, given that the accelerated expansion of the universe keeps making things get further away." It's a large, but finite, sphere. It's actually smaller than the Cosmological Event Horizon [0], which is the cutoff point, beyond which light can never reach us (again, because the accelerated expansion of the universe is effectively putting space between us and the event faster than light can travel) - estimated to be ~ 16 billion light years in radius.

If by "universe" you meant "observable universe" - that's fair, but still no. How long could humans conceivably last before we go extinct, or change into something that can't be called human anymore? I feel like 100,000 years is an absurdly high estimate, but let's got 10x higher.

If we figure out how to travel at c/2 soonish, and we say going somewhere counts as "exploring" (no need to report back to any one), then we have a ~500,000 light year radius sphere to explore. That's ~ 225 billion stars. For the near ones, we can shuttle back and forth, but for the far ones, that's a one way trip. So let's say 100 billion space ships. And that's if we leave tomorrow.

c/2, 100 billion space ships... If we start paring these back to reasonable figures, you start to realize: our species will barely scratch the surface of the observable universe. In 1,000,000 years, I think, best case, we can cover a 50,000 light year radius sphere. That's 0.000000000000003% of the observable universe.

[0] https://en.wikipedia.org/wiki/Cosmological_horizon

I don't know why it isn't as popular, but CatBoost should be on the list too https://catboost.ai/

The article recommends RF for tabular data because it is easier. In general I agree, but newer tools are making NN for tabular data as easy as can be... see, for example, fastai https://docs.fast.ai/tabular.html

Surprisingly, there are a lot of good tips in the US Gov's Plain Language Guidelines https://www.plainlanguage.gov/guidelines/

This is not a bad pitch

Regardless of similarity to Wolfram products, this looks cool. I feel like I could be in the target audience, but the documentation is quite sparse. I feel like

Your data ends up being represented as nodes in a graph that can be traversed by comparing a query (which is also represented as a vector) to nodes from the graph and use their cosine angle as a guide as to what node to traverse to next and when to stop.

Could be... at least a page, with some diagrams. Each node... is a vector? Has a property that is the vector?

plus then you lose voice typing completely which is not acceptable to me

I use GBoard on iOS, and on the iPhone XS with iOS12, you can still use the native voice typing while using GBoard

Aren't Jupyter notebooks R notebooks? Jupyter stands for "Julia, Python, R" I believe

I'll be honest, it's been a long time since I was studying classical mechanics, but 15 years ago I remember answering a similar question with "read the Feynman Lectures Volume I!" They are the edited version of the intro physics course he gave at CalTech... Here is a link: http://www.feynmanlectures.caltech.edu/I_toc.html