omegalulw

> AWS provides extreme flexibility. This really only becomes valuable once you hit a scale that you can optimize your infrastructure around the shape of your needs.

This is upside down? When you hit the scale where you get value from modelling your infra around your business use cases that's when you go on-prem or use services that let you rent bare metal. Before that you use cloud.

This doesn't make sense. The main incentive to use Blockchain for payments is that it is decentralized. If that's not a priority, and it's not for governments and big banks, Blockchain doesn't get you much. In fact, the cost of mining is downright harmful to the environment and wouldn't scale to billions of people using it for their daily needs .

On my mobile, for [1] KaTex loads in 900ms, MathJax in 1200-1600ms. Good enough for client side usage.

> I think you're an above average developer working at a below average company.

This. I'd conjecture that this has led to OP becoming overconfident.

I also think that OP should not be a manager. If one dev is doing 70% of the work then that's either bad hiring practices or micromanagement where other team members are not trusted to deliver complex projects.

How is the choice between fp16 and fp32 made? Is it like if any gradients in the tensor need the extra range you use fp32?

I think it's also important to highlight commercial USB webcams, like the Logitech 922. They are kinda expensive but still cheaper than a DSLR and will get you all of the low hanging fruits over webcam quality. DSLR would get you benefit on top of that but arguably, given that 720p is the common webcam resolution, you wouldn't be missing out on much. Except maybe features like improved focus and depth of field effects.

> Now, the obvious thing is N-squared and clearly terrible, but let's do it anyway and see if we can get some insight and improve it..."

By "let's do i anyway" do you mean to implement it? That's a terrible strat. You should not be touching the keyboard until you have an optimized or at least final design. If coding up the naive algorithm gives you some "insight" that you don't expect to have seen by thinking through the problem beforehand, you need to invest more in logical thinking.

This applies to both interviews and to your normal work.

It's weird because this is violating a fundamental principle that people generally associate with crypto: if you know the keys, you own the crypto. At least this is the case for Bitcoin.

Certainly on dedicated charging stations?

My understanding is that Displayport is already packetized, having USB4 packetize it again seems suboptimal? I think both Displayport and HDMI have official specs that work over the USB-C connector.

> Counter point: This is clearly not 100% true since I would not say the same thing about Haskell even though arguably I could.

This is not really a counter point. Any dev who struggles to write code in any modern programming language (after ramp of course) is arguably not a good programmer. Programming is far far more than code.

Can someone tell me if this is a good time to invest? Or would it go lower?

Just looking to invest in VTI, nothing else.

How is backing for stablecoins not regulated to kingdom come yet? Seems so asinine?

This doesn't well thought out either. What if you need to send larger amounts of money? Sending out money to everyone works when it's a reasonably small amount.

This is a terrible idea. Mistakes of this magnitude are often the result of multiple people making mistakes - so blaming a specific person is wrong.

Not to mention that this would force a culture of attributing blame to individuals which is toxic, encourages people to hide mistakes and when mistakes happen the first thing everyone would scramble to is to find a scapegoat. This is objectively worse for both the company and the shareholders.

The right approach for the government is to appropriately penalize the companies. The right approach for companies is to put in processes that make sure no single person is even in a position to make mistakes like this.

I don't know about other people, but finding something to watch on Netflix is frustrating, there's so much poor quality shows. The recommendations are not good (tbf that's most streaming services) and there's no option for granular filtering. For example, I hate dubs, I would like to pick shows where the original audio is some particular language, I don't get why that isn't there?

In contrast, on HBO Max / Disney+, usually there's far less content but since the quality is better you can easily pick something to watch.

> Absolutely! Because as we all know, everyone speaks English.

I understand the sentiment here, especially as English is not my native language, but for many domains, for anything close to the state of the art, English is lingua franca.

One thing of note is that there isn't a single format that's optimal for storing tabular data - what's optimal depends on your use case. And if performance doesn't matter much, just use CSV.

As a simple example, column based formats can significantly speed up queries that don't access the full set of columns. They can come in handy for analysis type SQL queries against a big lump of exported data - where different users query the subset they are interested it.

Other than the reasons already posted, I find it much easier to read code with 80 char line limits vs longer. For me it's like the benefit of reading an article in columnar format - going down vertically just feels more comfortable.

> to chase real genuine truth yourself. The ability to develop your own mental model of how you think things work, and your ability to probe & validate & make inquisitions into each little step: that scientific minded exploration makes all the magic happen.

Wonderfully put; this cannot be overstated! And I think you can apply this to most of math, physics, electric engineering, etc too.

One of the core benefits of C++ is that it is a high level language that's still super close to the hardware. Switching from C++ to other languages, if you have mastered C++, should largely be a question of "what things are automatically done for me" which is very different from "what are these new things I need to do" that you have to deal with when you switch between other languages.

For those wondering how this works, n & n-1 zeros out the least significant 1. Simply by doing this repeatedly you count the number of 1s.

I understand the sentiment but you should be judging the PR, not the source. Ask yourself: would you have happily accepted the same PR that the bot sent if it came from a human?

By all means, I am not against having bots identify themselves properly, my point is that "effort from bot PR is ~0", "it advertises their platform" are simply not the right reasons to judge this situation by.

It's not necessary to have p(0 <= v < 1) = p(0 <= E < 1). Only that P(f(X)) is uniform.

But this does bring up a good point. H(X||Y) = H(f(X)||f(Y)) for any bijective f if the distributions are discrete. When they are continuous this is not true, even with a bijective f. For example f = x^2 doesn't work even though it yields a binary distribution. Interestingly however, affine transformations work.

+1. The commenter above also wanted cared about bijective mappings, and squaring a random variable in [-1, 1] is not bijective. Squaring a random variable defined over positive real numbers would lead to a bijective mapping and the distribution would still remain uniform.

Actually, I find it hard to come up with a bijective mapping that leads to a non uniform distribution that's useful for anything practical.

> I thought entropy (in the Shannon sense) was a property of discrete and finite probability distributions. It's essentially a measure of how random a sample from such a probability distribution is. Notably, continuous probability distributions don't have meaningful entropy (or in some sense, their entropy is always infinite).

True, but for continuous distributions you can use the KL divergence against a uniform distribution :)

> During a hostile takeover, someone buys 51% of the shares. They then elect a board of people who will approve whatever they want.

This is confusing. During the TWTR discussions, one that regularly came out is the importance of a board's fiduciary duty. Buying 51% doesn't let you take the company private, and merging at 50% of the price seems like a monumentally indefensible decision, especially when it is with the 51% holders company. I find it hard to believe you can get away with that.

What about bcrypt?

> A real concern right now is that the course material in both academic CS and academic SE is behind by 5-10 years in terms of tooling relevant for industry jobs

Meh, the most important skill a computer scientist or a software engineer needs to have is to be able to think critically which good courses, especially ds/Algo and the like, will teach - tooling changes every month and varies across companies.

> We ran a campaign before Christmas were we were selling a new product that was only marketed on Facebook, we are certain that (almost) all customers found it though that Facebook campaign.

I think 50% may not be that bad. I imagine a lot of people simply search for a product directly when they see a compelling ad rather than click through (especially since I use ad blockers which sometimes block these clicks).