cwp

Yup. Couldn't read much of the article, but this is me.

I was in SF because employers demanded it, because investors demanded it. Now my company is not only allowing Bay Area employees to leave the state, but hiring remotely. Hell yeah, I'm gone.

Lots of little things that suggest Shor is so deep into politics he can't remember what a regular person is like. For example:

  The single biggest way that highly educated people who follow politics closely are different 
  from everyone else is that we have much more ideological coherence in our views.

  If you decided to create a survey scorecard, where on every single issue — choice, guns, unions,
  health care, etc. — you gave people one point for choosing the more liberal of two policy options...

He completely misses the fact that the liberal-conservative spectrum we see in American politics is pretty arbitrary and has more to do with tribalism then ideological coherence. For example, it would be "ideologically coherent" to be both pro-life and pro-gun-control. Or concerned about climate change and pro-nuclear-power. Or anti-racism and anti-immigration. So his conclusion is wrong. Highly educated people who follow politics are not more ideologically coherent, they're more ideologically conformist.

Another example:

  At some point, you have to be like, oh, actually, these people were motivated by racism. It’s just
  an important fact of the world ... 

Who's he talking about? Obama-Trump voters. He's looking at this sophisticated polling where they ask questions designed to probe for "racial resentment" and seeing no other way to interpret them. Ok... so why did they vote for Obama? Were they motivated by racism then? Did their motivation change between elections? These aren't rhetorical questions; I have no idea why Obama-Trump voters made those choices. But Shor is so deep in the numbers he can't see that his conclusion is nonsense.

Wow. This guy is a really strange combination of brilliant and clueless.

Is it really that hard to believe that iOS users do more searches than Android users?

Npm is very buggy for local dependencies - eg in a monorepo that contains node modules. That might be fixed in npm 7, but I doubt it.

OTOH, yarn handles this just fine.

If you're writing a compiler, sure, a hand-written parser is probably better than a generated parser.

However, I'd also make the related claim that we should be much more ready to write formal parsers for other use cases. Far too often developers implement a "pile of regular expressions" where a parser—even a generated parser-would be a better choice.

From the author's follow-up article [1]:

  How is this possible? Some of my planner colleagues will say it is density, but I've long rejected that simplistic 
  explanation. There is a lot more to it than a simple division problem.

He goes on to explain that the poorer areas tend to have smaller houses and narrower streets built on higher ground, all of which leads to lower maintenance costs. They were built with more frugality and risk aversion than the more affluent areas. This supports the thesis that Lafayette's predicament stems from extravagant infrastructure in affluent areas that wouldn't have been built without subsidies.

[1] https://www.strongtowns.org/journal/2017/1/10/poor-neighborh...

The problem with nuclear is that people are afraid of it. If we had spent the last 40 years trying to make it cheaper instead of trying to prevent accidents, it wouldn't be so expensive.

Clear violation of Betteridge's law.

I did a lot of work in Smalltalk earlier in my career, and it's still the best toolset I've ever used.

On one hand, yes, having the method as the unit of code is great. You just avoid all sorts of editing and versioning of non-semantic textual changes like you have in other languages. And it was really easy to look over your list of changes to see who did what when, or group them together to form larger patches.

On the other hand, collaborating with bare change sets was difficult. With no dependency tracking it was just a manually-maintained bag of class-and-method definitions. You could write those out to a file and import them into another image, but it was hard to make sure you had all the necessary bits.

This led to several Smalltalk-specific VCSs. The early ones were powerful, but basically required you to be connected to the net to do any work, because they were constantly communicating with a central server as you worked. Later ones were more snapshot based. I think Monticello was the first real distributed VCS - it preceded git by a couple of years.

What's interesting about Pijul is it gives you that "maintaining a set of patches" workflow that made early Smalltalk change management so pleasant, but also tracks dependencies between the patches so that you can easily mix and match functionality without anything getting left out.

The ultimate tool would be something that combines Pijul's theory of patches with Smalltalks semantic versioning. That would be amazing.

You could embed images using data: URLs. There are size limits in some browsers, particularly older ones, but you can manage some pretty high-resolution images under the limits in a modern browser.

Yup, Google Maps has better data.

But, on my no-longer-bleeding-edge iPhone, the turn-by-turn directions suck. It's constantly telling me to turn onto a street I just passed, or rerouting to find a way to get me back to the street that I'm already on. The worst part is that I can't trust it even when it's right. If it says "in 300 feet turn left," I have to desperately look at the street names because it's quite possible that I actually have to turn in 30 feet.

It was such a relief when I realized that it's not my GPS hardware, Apple Maps gives directions perfectly. And the map data is way better than it used to be. Not as good as Google Maps, but good enough.

Or even just stop lobbying the city to make it illegal to build houses.

Uh, you know that Apple makes iOS too, right?

Temporal is just the namespace. It has classes called DateTime, Date, Time, TimeZone, Duration etc.

Funny you should use the word "rational" here.

This is fascinating. It's a New York take on the Bay Area take on New York. It's the Media take on the Tech take on Media. It's illuminating, and frustrating. I keep looking for some reflection, some attempt to learn something from another point of view.

I'll have to read it again more closely.

I used the non-twin version, and recommend it to all my friends when they become parents. It changed our lives. Having a well-rested child makes a world of difference.

Nothing about this is convenient. Letting a baby cry and not providing comfort is emotionally shattering. Spending hours trying get a baby to sleep without abandoning them is exhausting. Caring for a sleep deprived baby is exhausting, physically and emotionally. Doing any of this without having enough sleep yourself makes it 10x worse.

The fact is, parents and babies both need to sleep to be healthy and happy. Most parents are trying to achieve that, and have to try a bunch of things before they figure out what works for their situation. Try not to judge.

No. We know people want this. Many existing customers joined the wait list to try it, and a few employees are using a ludicrously inadequate alpha version.

We might get the details wrong, but somebody is going to make a killing with this in the next few years.

Both non-competes and unions prevent competition in the labour market. Being against both is completely reasonable.

Trying not to get too excited, but damn, I've wanted to see this brought to life for so long.

I've had good results embedding a query language into the "query" component of URLs for RESTful GET requests. So something like:

  https://example.com/messages?posted-2020-01-01/content~Bob

would compile to SQL like

  select * 
  from messages 
  where posted < '2020-01-01' or content like '%Bob%'

It's a little bit constrained by the the characters that can appear unescaped in the query component of a URL, which is why it uses '-' instead of '<' for less-than and '/' for 'or'. But there are enough characters to allow a range of operators, and parenthesis is allowed for grouping. Also it's a super set of the usual way that browsers encode forms, so that '=' and '&' work the way you'd expect.

It works pretty well!

No, just the clip that Lou showed at the talk.

Yeah, I was at that talk. A little out there, but pretty awesome.

Yes! What's interesting about Gibson's vision is that he assumed connectivity would be hard, but VR would be easy. It's a world of payphones, fax machines, and "simstim" where you can receive someone else's sensory input and ride along inside their head as they move through the world.

That's not where we are now, but we'll be a lot closer to it if and when we nail VR.

Tell you what: you give me time and money, and I'll work on this program. Keep giving me time and money until I tell you I've succeeded.

In the case of the article, you are correct. I have a slightly different case where I'm wrapping scikit-learn model. We're NOT just calling another service and waiting for a response, we're doing computation, in Python. So the GIL is actually a problem.

Sigh. Yes. I have been there and done that (more or less) and it sucks. The root problem is that data scientists really want to use Python for machine learning, but wrapping a Python model in a service that uses CPU and memory efficiently is really difficult.

Because of the GIL, you can't make predictions at the same time you're processing network IO, which means that you need multiple processes to respond to clients quickly and keep the CPU busy. But models use a lot of memory and so you can't run all THAT many processes.

I actually did get the load-then-fork, copy-on-write thing to work, but Python's garbage collections cause things to get moved around in memory and triggers copying and makes the processes gradually consume more and more memory as the model becomes less and less shared. Ok, so then you can terminate and re-fork the processes periodically, and avoid OOM errors, but there's still a lot of memory overhead and CPU usage is pretty low even when there are lots of clients waiting and...

You know I hear Julia is pretty mature these days and hey didn't Google release this nifty C++ library for ML and notebooks aren't THAT much easier. Between the GIL and the complete insanity that is python packaging, I think it's actually the worst possible language to use for ML.

So... this thing gives full read/write/ddl access to production data ...via Slack?

That is a security nightmare. Hard pass.