Falcorian

My favorite quote on the subject is:

> Every great open source math library is built on the ashes of someone’s academic career.

From William Stein, lead developer of the computer algebra system, Sage: http://wstein.org/talks/2016-06-sage-bp/bp.pdf

Yeah, working from home for four years now and I only have one office and desk. It's used for work, of course, but also when I'm writing for my self, or playing video games with my friends.

I don't have any problem disconnecting from work, but it probably helps that I have a different computer for work that goes in a drawer at the end of the day.

I do follow points 2 and 3 though and I think those are essential.

It's definitely a culture thing, probably more at the company level but certainly country level culture matters as well.

At my San Francisco based fintech, the culture that's developed is:

For one-on-ones and small meetings where everybody is expected to participate, cameras are on.

For large one to many broadcasts (like an all hands) generally people turn the camera off.

But I would say probably 95% my meetings have cameras on, so I'm constantly "seeing people".

Yes, the models are not constantly learning. They only update their knowledge when they are retrained, which is pretty infrequently (I think the base GPT models have not been retrained, but the chat laters on top might).

False. Every time you transfer money from an account at one bank to another you're using ACH. Same with moving funds from banks to stock brokers or brokers to banks.

The real reason it doesn't compete is its slow, and many banks have terms of service that say something like "You can only set it up between accounts you own" so sending to other people is often technically a violation of your agreement with a bank.

I trust Vacuum Wars for technical reviews of vacuums and vacuum robots: https://youtube.com/c/VacuumWars

Looks like their big robot round up is about 9 months old, but they have single model reviews as well.

Edit: A little bit on why I like this review channel:

A lot of gadget reviews I read nowadays about household products are like "I like this robot it did a good job cleaning. Sometimes got stuck but not too bad."

Vacuum wars is like: "I worked 500 g of fine sand into our test carpet and ran the robot for 5 minutes. It picked up 362 g of sand which is 32% better than the average for this category and the best of the current robots that I've tried. Now, on the artificial hair test, the rotors tangled after just 50g which is a poor result for this category... "

He's set up a testing suite. He runs all of the vacuums through it. You can compare the results. It's how reviews should be.

They prefer to have them at the end because:

1. You don't want these cars to get banged around a lot and maybe have their weight change.

2. So you put them at the very end where they take the least wear and tear.

3. But they're not allowed to be the last car because they often don't have brakes.

So you put them as the second-to-last car.

This answer based on combining two of the captions from the page:

> Always the "Second to Last Car in the Train"

> Andrew B Baird: No brakes so it is not permissible to be the last car.

And

> The scale cars are now more robust? I notice that this car is not on the end of the train.

Because the same two numbers (routing and account) are used to deposit money AND debit money.

So if you post your account numbers I can pull all your money out. Now you'll likely get it back if you file a fraud claim, but that's an extra Hassel, and your out finds until they give you a provisional credit.

It depends on the company and the role. In general, I would be prepared for some white boarding questions, although I've found they're generally "a little easier" than the ones the engineers whine about.

A fun tech side of this post is that the random walk visualization is an SVG that is generated live on the page, and hence new each time. Very cool!

The code: https://lemoing.ca/scripts/knight.js

As a Physics PhD, I certainly went into grad school wanting to get a faculty position; is suspect most people did in my cohort.

Of course we're mostly data scientists now, because only 5% win the faculty lottery.

Data science wasn't a path that existed when we started. I wonder if its existence changes what people going into a Physics PhD expect.

I found the OSX specific XDG section helpful, because I've been setting it to ~/.cache for years! Good to know there is a better way that the OS understands (a bit a least).

Here is an overly simple example:

Pre-compute the recommended hotels for my top users every night. Now when that user comes back, they see a slightly stale recommendation, but it's lightning fast.

You can also pre-compute and cache some of the inputs cheer model, like maybe a vector representation of the description of a hotel.

That's my experience as well: apt is only "slow" in that sometimes it has to download and write gigs of stuff to disk.

Now brew and conda, they often take minutes to solve for the environment, which is what I think of when you say "slow". That part is probably mostly algorithmic, and it should be fast, unlike network and disk.

Exactly.

At my company, Type A are called "Data Analysts", but at Google Facebook they're called "Data Scientists". Type B are "Data Scientists" at my company, but "Machine Learning Engineers" (or SWE-ML or some other combination) at Google and Facebook.

As a Type B, at my company, I'm on the same pay scale as the SWEs. The Type As are not.

Definitely, and I've added some of the same ones.

What I meant is it doesn't seem to parse the same range of formats, so it rejected some of Pi Hole's filters as malformed.

It is a nightmare! I just finished fighting with it this weekend.

The controller only officially supports ARM 32, and the mongoDB dependency only supports ARM 64.

In the end I I used a 64-bit Ubuntu, 64-bit mongoDB from the mongoDB repo, and downloaded the controller deb from the Unify site, which is apparently architect agnostic.

Really simple stuff, compared to the self watering plants, etc.:

- AdGuard Home DNS (Pi Hole was just not stable for me)

- Unifi Controller

- Torrent box (come get your Linux distros!)

- Always on platform to host various scripts on; one backs up my blog to Internet Archive when it changes, for example.

- Twitter bots that report the RPi status (just went down for reboot, etc.)

- Bot that watches Caltrain for delays and Tweets them to me.

- Platform to play around with orchestration, DevOps, etc. I learned Ansible using them. Still want to move it all to Docker, but that'll have to wait.

I've run into the same. It often would come back from a reboot without DNS restarted correctly. I never went through a Pi Hole update without having to manually fix things either.

A few weeks ago I switched to Ad Guard Home. Has been rock solid since. The big dancehaus it doesn't support the exact same filter files.

They are almost certainly not allowed by their employment contract.

Confusing, yes. In this case, it's GeV.

Similar story with me: decided I didn't want to join the long-term job hunt and went to Insight (4 years ago).

I've been very happy doing data science ever since. It's great to work more collaboratively, ship more quickly, and learn from great engineers.

ESC is the key that (roughly) takes you back to normal mode from insert mode, visual mode, etc. If you aren't using it, then you're either using one of the other ways to get back to normal mode (like CTRL-[), or you are using Vi in a very non-standard way.

See, for example: https://en.wikipedia.org/wiki/Vi#Interface