Random_ernest

I recall a story where a friend was unable to publish a paper in which he wrote an alternative to a very commonly used commercial tool (that virtually everybody used) with roughly 10 times better performance. He open sourced it and all, it was extremely useful, but there was no new methodology, it was simply very well implemented.

At a talk of his it lead to a very heated discussion where an older professor accused him of wasting government money on such nonsense.

Cold calling businesses is definitely not illegal in Germany if you got the contact info from their website.

The huge difference between spiritualism and religion is that religion is dogmatic and systemically organised. In my opinion freeing spirituality from the clutch of religion is the final step, not the other way around. So being spiritual but not religious is in my opinion not something that inevitably leads to religion, but if we do it right there is no need for oppressive religion anymore.

Bayes Theorem is one of the most fundamental theorems in the history of mathematics. I have yet to work in a field where it doesn't have deeply fundamental applications. In many cases expert knowledge or heuristic rules serve as prior.

Saying it is overrated is like saying sun or air is overrated.

> by automating consumption

It's called the subscription economy for a reason. Automated consumption exists and is constantly growing. The money spent on subscriptions grew in the US by >50% from 2010 to 2015.

Testing out the demo:

SELECT * FROM trips WHERE tip_amount > 500 ORDER BY tip_amount DESC

Very interesting :-)

From my very limited understanding the Graphcore machines outperform CUDA only significantly in inference, in training the improvements might not be sufficient to switch technology.

With likely very dire results, yes I think you should. If your mothers insurance rate goes up, since you got one of these dna tests for Christmas, she should be involved in the decision to publish this data in the first place.

But the problems where we apply human labour are vastly different from the ones where we apply machine labour. In (most) tasks where we apply human labour a few errors are tolerated.

Having worked extensively on synthetic data, what's your "verdict" on the topic? People seem to be very divided about it.

To me it's not about input being the bottleneck, but about Vim having a modular grammar, such that my brain can stay in "problem space" instead of "how do I edit the file space".

It allows me to have a deeper level of focus.

Very interesting. I feel like I have written this exact sentence somewhere online only with the languages reversed. Tried to get into FP multiple times with Haskell, never clicked, then tried Clojure and felt productive after 2 days.

For some years I was the developer you talk about, who always knew everything better, now I am a manager at a startup. So I was able to see both sides of the discussion.

I feel like it always comes down to a communication issue. Team members need to feel heard, and feel acknowledged that you fully understood what they are suggesting. On the other hand they need to understand that there are constraints that force a decision that might not technically be the best, but is pareto optimal. When this discussion occurs I sometimes bring up the analogy of these space movies where they put everything on a table that the astronauts have up there and try to make something out of it that mostly consists of duct-tape. Sure there are better tools, but the stuff on the table is all we have. Another thing that took me very long to understand as a developer is that innovation is a risk that might not always make sense economically. A thing that took me long to understand as a manager is that the happiness of your devs is one of your most precious resources that you have to manage well.

> Even the example of hosting complexity being replaced by cloud companies seems kind of silly to me. Maybe that’s saving very small companies a sizable fraction of their engineering resources, but I really doubt it for medium or larger companies.

This might even lead to an _increase_ in demand for software engineering, since now small companies can write their own custom software cheaper and more reliable. It's called Jevons paradox.

I just would like to add to all the great comments, that clojure also has access to the python ecosystem via libpython clj.

One could even argue that a lot of startups are VC backed R&D. Only VC does not care for the scientific, not-directly monetizable outcomes.

I haven't figured that part out either ;-)

From what I understand, you asked four questions in one: 1. how do you learn, 2. How do you decide what to learn, 3. How do you manage your learning time and 4. How do you handle the pressure of that huge mountain of stuff you don't know. I will try to break it down a little. My Job requires me to rapidly understand fields I've never worked in and try to understand as much as possible in very little time, so I feel like I can contribute to answering it. I am not saying my thoughts to this topic are the best TM, or particularly well thought through, but it works for me.

1. How do you learn: My learning strategy might seem a little weird but I'll explain it anyway. For me the first part about learning is about familiarity. If your brain sees to many words it does not know it subconsciously shuts down and you get frustrated/demotivated (at least for me). So you have to iterate over a topic in order to feel familiar with it's vocabulary. Just think of those Wikipedia rampages where you go deeper and deeper down certain words until you don't know where you originated from: that's because you are not familiar with the vocabulary of the field. So my first step is to learn the vocabulary of the topic by 1: reading a short book about the basics (or the first 100 pages or something) and 2: I find a place (online) where people working in this field hang out (e.g Reddit, HN etc.) and just read what kind of problems they have, which kind of words and tools they are using which kind of projects they work on. I do this daily, multiple times, and follow on things that really spark my interest and try to understand as much as possible. My brain works very interest based. I try to answer simple questions in forums, and try to get involved but only in simple stuff (since obviously I am just learning). At this point I try to apply the very basic things i've learned and iterate myself further by asking basic questions about stuff I don't understand etc. After having reached a certain familiarity with vocabulary and basics I try to explain why does tool X exist, what problem does it solve, what pros and cons exist. I think about how I would explain the necessity of X to someone who is not in the field. After that in the second phase I learn mostly like everybody else from books, online resources and just doing what I learn. But now I can read books a lot faster and with much fewer frustration because my brain is familiar with it, knows why X exists and which cool projects X is applied to. Thinking about it, my strategy is mostly tricking my brain into not being bored or overwhelmed.

2. How do you decide what to learn: I learn because my job requires me to be familiar with Y. That's one part, can't really change much about that. The second part, for me personally is purely interest based. I always try to learn concepts/methods instead of tools. Don't care if it is useful or you will ever really apply it, BUT: if I learned a useful concept instead of a tool, I will always be ready to apply it somewhere else. We humans are masters in generalizing things and applying concepts we have seen somewhere else.

Since you asked about HYPE-TECH-A vs something that truly interests you: I always in my life picked my interests not the hype, and it always worked out. If you are motivated to learn something you can gain knowledge several times faster compared to force feeding yourself something that you might apply maybe somewhere in the future.

4. How do you handle the pressure of that huge mountain of stuff you don't know:

Just have a good mental health. Be aware that staring to long into the abyss of stuff you don't know will never lead to anything good. You have to be aware of the things that you don't know, but let it give you a joyful humbleness instead of fear. Just think about it this way: you will never run out of interesting things to learn. The joy of learning will always be available. Your mind is not a commodity of your future employer, instead learning new things should be your privilege and bring you joy.

Probably not the question you had in mind, but what Lisp and what made you choose it?

AWS has roughly double the market shared. So it is very likely that they take that title. https://www.parkmycloud.com/blog/aws-vs-azure-vs-google-clou...

Thanks for your honesty, I thought I was the only one.

> The higher the frequency, the steeper the gradient.

Losses also scale with frequency.

And while other acceleration principles have higher voltage gradient, they still can't reach TeV of collision energy, since you either can't concatenate them, or you can't control the beam sufficiently well.

As I wrote in another comment, stuff like plasma wakefield accelerators need to get better by a factor of 1000 before they reach the realms of SRF accelerators. This wont magically happen over night.

And most of these people have tenure, they hardly care for trends.

I have been to all major accelerator physics conferences over nearly the last decade. Despite clic there is no one seriously claiming to go beyond LHC energy without SRF. No one. Not the plasma wakefield guys, not some strange theoretical accelerator principle. No matter what you read on Wikipedia.

Your namedropping comments with very little actual content, disregarding the work of thousands of scientists show a certain arrogance that is super annoying.

That is simply not true, the plasma wakefield accelerators are really a rather new technique but they are NOT meant to be used at such high energies, but rather for small experiments. Also (as far as I know), while they do have vastly superior energy gain per meter, you simply can't concatenate several of them (as you do with SRF cavities). So at the current state, if you want to reach TeV of colliosion energy, SRF is the only way to go.

Current plasma wakefield accelerators are in GeV range, we need TeV.

That is really vague. Even if there are interesting advancements in photonics, I seriously doubt that it will be able to accelerate particles to terra electron volts of colision energy in a controlled manner (in the next few years). Everybody I know that works in accelerator physics believes that for large machines, superconducting radio frequency cavities are the way to go.

A little off topic, but is there a german version of Solaris that is "readable"? The one I got reads like it was done in google translator (back when it really sucked).

I think people severely underestimate the scale, complexity, and time it takes to build large scale accelerators. There wont be some magical new accelerator popping up. Building them takes years, sometimes decades. Additionally you need a lot of experience in building, operating and designing those. Having worked in accelerator physics for quite some time my opinion is that China does not have the skills to do this. Research labs that built such machines had decades of prior experience and the community support. China has neither of those.

Imho the next big accelerator will be either some LHC upgrade, clic (very very unlikely) or the ILC.

A senior scientist told me they grow mushrooms now in that tunnel ;-)

What you are referring to is used in e.g. CAD modelling, we call them non-uniform rational B-splines or NURBS.

Recently there has been super interesting research on using these nurbs as ansatzfunctions e.g. for FEM stuff. They got a lot of applications.

In the Hyperion books by Dan Simmons there is a being that can move in higher dimensions and the books will reportedly get a movie. Maybe they will adapt some effects like this.

I took it from the article that dropping requests was permitted (since that happens when all servers go down). So my assumption is still that a better solution is that the load balancer allows only a specific number of requests per server and rejects or caches the rest. I would even argue that requests being rejected is more understandable for the user than the website simply not being there for a certain time.