Seven Year Exercise Windows

Most startup option grants come with a 90-day exercise window when an employee leaves the company (voluntarily or not). This is standard practice. Essentially the problem is that this can force employees to take a major personal financial hit to exercise their options when they leave or are fired, and perversely this is worse the more the company has grown during their time there. (Exercising options comes with a tax hit proportional to the amount the stock has increased in value since the option was granted.)

There’s an easy solution: when an employee leaves, convert their ISOs to NSOs and extend the exercise window.

Today my employer TrueAccord extended our exercise window to seven years for people with two years or more at the company. I’m proud of what this means — this is a progressive move and it shows a dedication to the employees of the company. In general I’m skeptical about equity as a carrot for startup employees (see here for an extremely skeptical view), but this is the right way to do it.

For more, see
https://github.com/holman/extended-exercise-windows
and https://zachholman.com/posts/fuck-your-90-day-exercise-window/
and point (2) in http://blog.samaltman.com/employee-equity

AI Clears New Hurdle; beats European Go champion 5-0

Today Google’s DeepMind team announced that they built a Go-playing AI which beat the European Go champion 5 games to 0. This has been a long time coming! Look at timeline of games conquered by computers:

tic tac toe: 1952
checkers: 1994
chess: 1997
go: 2016

Let’s make a simplistic assumption that the game-playing ability of computers varies as Moore’s Law, which means that it doubles every 18 months. Software has improved as well, so this is a lower bound

So beating checkers required a computer 2^63 times as powerful as beating tic-tac-toe. And chess required a computer 2^4.5 times as powerful as checkers, a factor of 22.6. But the Go program was 379625062 (or 2^28.5) times more powerful than Deep Blue.

These differences are shockingly large, and it turns out that checkers and chess are virtually indistinguishable on the spectrum of human cognition. Both are about 70% of the way between tic tac toe and Go — but who knows how far the spectrum goes?

Amazing.

Political Sunday Morning

Some links that will fill you with outrage:

Police Carelessness Wakes Local Citizen

Tuesday morning I was woken up at 5AM by a police siren. As I lay awake, I started thinking. I couldn’t be the only one woken up. Should police cars use their sirens at night? It seems like a classic example of diffuse costs and easy-to-see benefits. But is it worth it? Let’s make a quick back-of-the-envelope calculation.

The cost of running a siren at night can be modeled by the following equation (I love Fermi problems!):
LengthDriven * (2 * SirenDistance) * Density * PctWoken *
DailyIncome * ProductivityLoss

where:

  • we start with the LengthDriven with the siren on
  • multiply by twice the SirenDistance, how far away on each side of the police car a person can hear the siren, to get the geographic area affected
  • multiply by the Density to get the number of people potentially affected
  • multiply by the PctWoken to get the number of people who were woken up
  • multiply by the average DailyIncome to see how much economic value those people create each day
  • multiply by the ProductivityLoss (as a percent) that they experience when groggy to see how much economic value was lost by the siren

This model makes some assumptions. It assumes the police car drives in a straight line, that the density is uniform, that the PctWoken is constant within the SirenDistance and zero outside of it, that everyone works (so ignoring children), that everyone works a day shift, that the ProductivityLoss is independent of DailyIncome, etc. But it seems like a reasonable first step.

Let’s plug in some values:

LengthDriven = 0.25 miles

SirenDistance = 400 feet \approx 0.08 miles

Density = 18,187 people/square mile in SF (from Wikipedia)

PctWoken = 5% of people. I made this up out of nowhere.

DailyIncome = $45000 yearly per-capita income in SF / 200 work days per year = $225 per day

ProductivityLoss = 25%, I made this up too

This gives us an economic cost of $1023 every time a police officer flips a siren on at night.

Even if this only happens once per night in SF, it creates a cost of $375,000 over the course of the year — equivalent to the salary of about 4 police officers, or about 1% of the SFPD’s budget. Use of sirens also appears to be dangerous. I wonder what the benefits are — how much additional public safety is provided for this cost?

I reached out to the SFPD asking if they have any guidelines about siren use — stay tuned.

What is an order of magnitude?

I’ve always wondered exactly how to define an order of magnitude. At ideas42 I had a particular colleague with a PhD in Econ; one time I mentioned that I think use it in a “fuzzy” way, and he responded that he always uses it in an exact way.

So what is the exact way? I think there are actually two ways to use the phrase. Is it a property or is it only a relation?

  1. Property: every number has an order of magnitude that is equivalent to its power of ten (so the OOM of 153 is 2 because 153 = 1.53 * 10^2). This implies a relation: two numbers have the same order of magnitude if each of them has the same order of magnitude, and you calculate the difference in their OOMs by subtracting one from the other.
  2. Relation: two numbers x and y, with x < y, differ by N orders of magnitude if

    (x * 10^N) < y < ( x* 10^(N-1) )

Notice that these two definitions have different “predictions”! Example: is 153 an order of magnitude more than 53? Definition (1) would say yes, but Definition (2) would say no.

The Wikipedia page actually uses both definitions without trying to explain the contradiction — the first sentence is a statement of definition (1) but the John Baez quote is definition (2).

The general principle is that objective-sounding, mathematical terms get used for rhetoric even when they are loosely defined; dig deep and nothing is as precise as it seems.

How do you use the phrase?

Architect Donald MacDonald

MacDonald is an old man with ( I assume) an illustrious past — his firm designed the new span of the Bay Bridge, at least. And in a few weeks he has the grand opening of a bridge in Portland.

His one good, well thought out big idea is 3D zoning. Right now in most places zoning covers the lot– but shouldn’t ground floor spaces have different rules than 4th story? I wish he had spent more time on that.

This talk was laid out (he claimed) starting from lowest-income housing and moving upwards. Really this meant 2 minutes of a box he designed for homeless people, three camper/RV style homes he designed, and then a bunch about his so-called “suburban villages” and “urban villages”. Ultimately I think he’s an architect of small things rather than big ideas. And his small thoughts are fascinating. In particular he pays a lot of attention to lines of sight, though he never summarized it in quite that way. But he talked about living spaces on top floors so that you see the good part of the tree rather than the trunk. And in the villages he’s designed, he angles nearby houses in such a way that you can see through gaps between other houses into the distance instead of just looking at the houses. And in the mixed residential-commercial buildings, 2-3 stories with the commercial space on the ground floor, he talked quite a bit about how he put the planters in such a way as to provide privacy on the deck.
The trouble came when it came to delivering on the title of his talk. “Democratic Architecture” sounds like a big idea. “Practical Solutions” was more his strength, but not to “Housing Crises”. Essentially his vision of housing is much the same in cities and in suburbs — small, self-contained, mixed-use “villages” about one city block square. A good example is this development in Oakland though he also showed a number of others. He talked quite a lot about affordability but seemed to think that most of the reason housing is so expensive in San Francisco is because of needlessly expensive building practices, rather than the price of land itself. This is why he thinks that the building height cap in San Francisco is no big deal. To support this, he mentioned a friend of his in London who was able to find a cheap apartment after looking at 100 others, seemed to me more like a counterexample. And he talked about some admittedly cool work he did a few decades ago in San Francisco where he was actually able to find decent lots that were undesirable by conventional standards (around Duboce Triangle).

Anyway, interesting issues. I asked a question about designing for homebuyers vs architects — since most of his talk was about homebuying (e.g. his units are designed to be expandable as you get more income) which he completely dodged. He also talked almost not at all about household size, which is important given that much of what he designs is small by conventional standards (14×17 including a lo
ft bed, in one case). He mentioned once that “asian” families “like” to have a lot of people in one space.

Finally, a small thing. He mentioned that it’s actually easier to build tall things on the hills of SF rather than the low-lying areas
because that’s where the bedrock is! And the Financial District is basically mud.

Get Broader

Convene five of your friends who collectively have some expertise in each of the following:

  • Physics
  • Chemistry
  • Medicine
  • International Politics*
  • Literature
  • Economics

Every two months, read one book / paper by each of that year’s Nobelists.  Meet and discuss. Make the economist wear a special badge.

*For the peace prize, of course. I couldn’t figure out what it means to be an expert in peace.