Purpose

• I have a mailing list that I started 6 or so years ago.
• I have never posted on it.
• I want to start using it to bounce ideas and start a community.
• I don't want it to be jarring, so I want to write a first post.

Outline

• Hey, it's been a while
• Life Update
• Intention for newsletter
• Here is some stuff I’ve written, and here is a list of things I’m thinking of writing about. Would love to get feedback. Join my discord.
• At first will be single mailing list, but I’ll be trying a few things and if some people like it and some people don’t, then I’ll split them into multiple lists (articles from blog, personal updates, etc)
• Cancel subscription link
• Links to things I'm thinking about recently

Risk-adjusted expected returns for being an early engineer at a breakout company…
Time > money - aim for high dollar-per-hour and learning opportunities, not high income (unless you are learning something you would be learning anyway)
Pareto Principle applied to interviews, directed practice…
Spaced Repetition (particularly for algorithm problems) - you can continuously practice interviewing for your entire career.
Interview with places even if you don’t want a job.
Calibrate your market value
Give you leverage at your company
“Competence triggers”

Purpose:
Tell a story about Knuth (it’s kind of cute that he didn’t like emailing, and some of it is funny)
People have said this is worth more than my PhD, but I haven’t discussed it.
I found an error in Knuth’s The Art of Computer Programming vol 4A
Sorting DNA fragments was a bottleneck for building a de Bruijn graph for DNA sequencing. There was a paper that defined a hash function that hashed k-length strings where the positions form a de bruijn sequence (an euler traversal of a de bruijn graph, so essentially a DNA assembly).
The paper was binary only, but Knuth’s book expanded it to larger alphabets (e.g. DNA) the algorithm slightly better than in the paper, but the math didn’t work out and I had to use GCD, so it wasn’t O(1).
I’d like to explain the math here and why it didn’t work out, and how I used GCD instead. This will take the majority of the time to write the article though.
I emailed Knuth to ask about this, and he wrote a letter back to me (scan this)
Later he wrote to me again saying that I was right, and that he has since corrected it, and I’d get the fabled Knuth Reward Check (link)
I was moving to a different house, so I sheepishly wrote back to say thanks, can you please send it to … instead? But of course this went unnoticed.
Always said I had it, but never actually had it. Academic fraud.
When I visited the US again in 2016 someone suggested that I swing by Stanford to pick it up, but I didn’t think it would have been appropriate for me to bang on his door and shout “Donald Knuth, you owe me money! Where’s my $2.56?!”
This was in 2014. In 2020 a stranger emailed me to tell me “Donald Knuth is trying to get in touch with you”.
Show photo of envelope with bounced check (yes of course I kept everything Knuth has ever written to me)

• Identifying impactful Avoidance Areas along a route
• Overlap aware Avoidance Area ranking (for an avoidance area set)
  • How to do a stack ranking analysis when the statistics overlap?

Routes are important for UX
ETA, the way a route looks, the pickup/dropoff location (this part is huge - could be interesting to write a post about this by itself)
Another post idea: How routing for a self driving car is different
Lane change probabilities
Dynamic avoidance area and lane features
Time based routing (cite the recent book)
Some lanes are only available at certain times.
Balancing Risk and ETA costing (and keeping those updated)
Fleet distribution
Making this behavior accessible in simulation, on the road, in python notebooks, in our tools to edit simulations, in our debug tools…
Geofencing
Determining pickup/dropoff locations during different scenarios.
Handling reroutes.
Avoiding sink lanes (where we are forced to stop until we can lane change)
Need to analyze them
Most analytics and BI people know Python, not C++...
How I wrapped our router code into a Python library
How it was done: why extern C, why shim functions, etc.
The lessons learned, what I would do differently.

Why fungibility is necessary for a currency, why bitcoin is not fungible, why I am concerned about this, what to do about it.
Benefits: sethforprivacy.com will read/post it, authors of bitcoin privacy tools probably would too. I have explained this multiple times (if you’ve said it twice, write it down - Jack Butcher). Gives me an opportunity to paint my vision of crypto as symbiotic (talk to Paul Sztorc about BIP 300).

https://alexbowe.notion.site/Fungibility-e903e5f2b3334dd18a2eae4757bafad5

• My second and favourite paper from my PhD - the mathematical patterns are so beautiful and the process was really enjoyable.
• I was in a slump in my PhD, enjoying my time in Tokyo too much. My professor from Melbourne was in Tokyo and we caught up. He invited me to Helsinki to work on a paper and hang out. I was there for a month - 3 weeks of him showing me wine bars and teaching me how to roll cigarettes, but we wrote a paper in the last week
• De Bruijn Graphs (link to previous post) are used in lots of bioinformatics applications, but they use a fixed-parameter, k, which affects the quality of the results, and the best value is different in different areas of the graph. To get around this there was research that iterated over this and built many graphs.
• We designed a data structure that represented all values of k and could be changed on the fly, in a single graph, with minimal space overhead (2 bits extra per element).
• This was later used to improve short-read DNA assembly to be as good as sanger sequencing.

The original metaverse (swyx.io, micro.blog). Webmentions (backlinks for the web).
What Roam could be. What Twitter could be.
Broadcasting, information theory (peace, love, information theory).
Distributed Search over Combinatoric Idea Space.
Serendipity. Most good things have come from this. Outbox Infinity > Inbox Zero.

Idea:

• Outbox Infinity - publishing articles. Permissionless.
• Inbox Zero doesn't actually do anything except reduce a number on a badge.
• Inbox Zero: Other people are adding things to your todo list in a permissionless way. Fuck that.
• Reaching out to people to say thanks for something they wrote.
• Outbox Infinity > Inbox Zero

Related to #8 .

Building a Dyson sphere to capture output from your every-day life.

Readwise, #9

What is Incremental Reading
My ideal implementation
How SuperMemo and Anki implement it and where they are lacking.
Incremental Writing

Part of the application process for Maven.

• Whether I would recommend it.
• Advice to PhD students.