One of the better decisions I ever made was not ignoring emails from folks recruiting for Amazon in 2003. It was a different place back then, and truly like "the world's largest startup." The other thing that was different back then was that the people I worked with were all so blindingly smart. I'm pretty smart, but in 2003 I often felt like the dumbest person in the room[1]. In 2025 that feeling was rare, and not simply because I had 20 years more wisdom.
I left Amazon for the third and last time a couple months ago and have no regrets.
If you're still there and reading this, Amazon still has a lot going for it as a place to work. But it's not the electric place I recall from 20 years ago. I'm not sure if there is any company that can match both the startup-like freedom of action with massive scale of early 2000s Amazon.
[1] In case it isn't obvious: this is a desirable condition because it means you get to learn something.
I never worked for google (interviewed but was rejected) but I have suspected this is true -- that there was a sweet spot where google was a magical place, but that time has long passed.
I don't necessarily need an electric place to work. Smart people and a boring mission/tech stack is fine. I just think their layoff(s) in disguise when they forced everyone to RTO is/was a slap in the face. Instant, permanent loss of trust.
It's still easy to feel like the dumbest person in the room at Amazon... until you cut through the bullshit and realize everyone else in the room is a complete impostor skilled in maximalization, social engineering, politics, and nothing else.
There's a bunch of milestones, for me the standout one was managers starting to abuse marking tickets for large events as "secret" to stop people from reading their screwups. Someone leaked that the cause for some large AWS outage was someone oopsing some CLI command, and it seemed to trigger a pretty large shift.
I wrote a talk entitled How Humans See Data that puts several of these ideas, among others, into a coherent framework based on research by Bill Cleveland.
Thank you for posting this. It was an excellent, well-paced talk. 40 minutes flew right by.
The part about three levels of estimation was really eye-opening. Even though it's obvious in hindsight, I didn't make the connection until it was spelled out for me.
In 1999, front end templates were written in an internally developed macro language named catsubst and served by a C (maybe C++?) application named Obidos. Later (starting in 2003-ish), front end code was written in Perl (Mason) and served by another application named Gurupa. The transition to Gurupa was very long and arduous. I've no idea how the site works now.
Calling it an "internally developed macro language" is stretching it a little bit. It was an extremely simple blob of C++ code that scanned text looking for one of a very small number of known placeholders, and filled them in with the relevant value (session ID, user name, user email etc.) I suppose that over the years between 1996, the capabilities of catsubst may have been expanded somewhat, but certainly when it started there was no formal concept of it being a macro language. In fact, at the very beginning, we experimented with just using m4 (which is a formal macro language) but it wasn't quite right for the job.
Your control is greatly reduced in aerobars, but more importantly, your ability to defend your space is hugely reduced, which makes riding in a pack extremely dangerous. This video explains why having the ability to protect your bars prevents crashes: https://www.youtube.com/watch?v=sRIsWzrMmBs
Cycling has some dumb rules (e.g. sock length), but banning aerobars in mass start races is not among them.
This. Aero bars are nice for flats and descents. I wouldn't want to have to do a tough climb without drop bars or at least flat bars. And I definitely wouldn't want to ride in a pack with drop bars with my hands not on them. That sounds like a great way to wreck.
I'm like this, but when I have too much caffeine I notice that it's harder to turn the monologue off. It becomes a little shouty in fact. Brains are weird.
Brains are so weird! When I get deep into thought sometimes my internal monologue doesn't turn off but it sort of spinlocks by repeating bits of songs that are tangentially related to whatever it is that I'm thinking of abstractly. I have a sense in general that there are two parts of my brain that are semi-independent.
I play time-attack puzzle games; and some of my best times have been while I'm listening to an audiobook, music or deep in thought about some emotional problem. It feels like I actually perform both tasks better if I'm doing them simultaneously because my "vague processing" isn't interrupted by my "sharp processing" and vice versa.
When I have too much caffeine I've noticed that I have very strong ability to recall auditory information, specifically tunes. I get a sort of perfect-pitch effect where I can consistently recall music in the proper key and replay it in my mind. It also happens late at night.
How do you remember music? Can you try to describe how you recall it?
I have this problem with music/sound specifically. I cannot remember it at all. For example, I have favourite artist that I heard all of his albums 50+ times. If you play a random track of his to me I won't be able to identify it. Another example; My favourite all time soundtrack that I heard countless times, I cannot recall how it goes. If somebody asks me to "sing" the melody I will fail miserably. Oh and I don't know lyrics of any song. Any. I tried to learn though...
I still get tired of listening to the same tracks over and over again :( .
You've got it more or less. The way I say it is that bad ideas can't hide in writing. You can fool yourself and others with a powerpoint or a spoken argument. But clear writing demands clear thinking. Poorly constructed arguments are (usually) obvious as such when put into writing.
In practice I find prophet far superior. Prophet can deal deftly with the kind of problems that show up in real-world data like missing values and radical changes in trend from shocks to the system you're modeling.
For textbook problems like forecasting electricity demand, sARIMA works just fine, but you'll struggle forecasting anything more complex. This is not to say that it's impossible, but it will require far more expertise.
I left Amazon for the third and last time a couple months ago and have no regrets.
If you're still there and reading this, Amazon still has a lot going for it as a place to work. But it's not the electric place I recall from 20 years ago. I'm not sure if there is any company that can match both the startup-like freedom of action with massive scale of early 2000s Amazon.
[1] In case it isn't obvious: this is a desirable condition because it means you get to learn something.