Jalex Stark |

I consider myself a *theorist* in the broad sense: I like to consider compelling-but-vaguely-specified problems, find formal statements that carry some of their essence, and then attack the formal questions with the tools of mathematics, computer science, physics, philosophy, etc. In my current work on algorithmic trading, I'm learning to add computers and data to this picture. This is slow-going.

I was formerly currently a quantitative researcher at Jane Street Group. I consider myself an effective altruist. Before that, I was a graduate student in the CS theory group at UC Berkeley, advised by Umesh Vazirani. Before that, I was an undergrad at Caltech, advised by Thomas Vidick. Here is a CV, updated August 2019. Here is my archived undergrad blog. Email: jalex at cs dot berkeley dot edu |

Personal Info |

My first name rhymes with the more common "Alex". In particular, my name is not pronounced like "Jay-lex". My given name is "Jacob Alexander Stark", and this is reflected in some of my official records. Etymologically, my first name is a contraction of my given first and middle names.

I don't like playing masculine roles or being identified as a man. If you refer to me with pronouns "they" and "them," I'll be more comfortable around you than if you refer to me with pronouns "he" and "him". I won't remind you of this unless we have a conversation about it.

I accept anonymous feedback here. I especially appreciate being told that some specific actions I've taken are likely to hurt people. I'm worse than average at noticing and interpreting social cues.

Research Questions |

I spend most of my days working on specific (proprietary) instances of the general problem "design and enact decision procedures that identify market inefficiencies as well as possible, measured in terms of maximizing the ratio (expected value in dollars of trading against the inefficiency) / (amount of human time required to find the inefficiency and execute the trades).

Accordingly, I'm very interested in questions about

*human psychology*, especially in the sense of biases that cause large deviations from expected utility maximization- deriving the behavior of systems of many agents from facts about the individual agents, a la
*macroeconomics*and*statistical mechanics* *statistical learning theory*, especially to the extent that one can design empirically effective ad hoc inference algorithms by using analogies to provably correct ones

**Outside of algorithmic trading**, I'm interested in finding structural changes to collaboration that accelerate the pace of scientific research. Some of my favorite things in this space include

- reinventing discovery, a book by michael nielsen about science in the age of networks
- bridging the gap between intuitive knowledge and formal knowledge, especially via
*writing machine checkable proofs* - stackexchange + mathoverflow, sites designed for technical professionals to ask and answer very specific questions
- the emergent ventures project by the mercatus center, and in general any attempts to use startup-like models for academic-like projects
- LessWrong and alignmentforum, group blogs dedicated respectively to general rationality and AI alignment
- Ought, a small nonprofit research lab, and specifically their factored cognition and dialog markets projects.

I used to participate in **academic computer science**. The following legacy bullets were written in that time.

Here are questions which I think are morally relevant and where theorists may be able to make significant contributions:

- How should one act in the presence of uncertainty about the merits of various moral systems? See e.g. William MacAskill's PhD thesis.
- Is it possible in principle, disregarding engineering constraints, to design an agent X such that (after an appropriate training period), X can accomplish your goals better than you can? See e.g. Paul Christiano's proposal for a meta-algorithm (IDA) for this problem. As a concrete question: can we find a toy model in which some instantiation of IDA provably works? Conversely, can we find a toy model in which there is a provable barrier to IDA-type algorithms?

My favorite "big" open questions inside of academic computer science include:

- The quantum PCP conjectures.
- Does QMA = QCMA? That is, for all problems with quantum proofs which can be efficiently verified by a quantum computer, does the same problem have classical proofs which can be easily verified by a quantum computer? (Resolving this question in an unconditional sense would separate P and PSPACE. I'm interested in evidence towards this question, say by stronger kinds of oracle separations than are currently known.)

Papers |

You can also see these on my google scholar profile.

- Trading locality for time: certifiable randomness from low-depth circuits (Abstract)
with Matthew Coudron and Thomas Vidick. Technical report, arxiv:1810.04233

*QIP 2019.* - Unconditional separation of finite and infinite-dimensional quantum correlations (Abstract)
with Andrea Coladangelo. Technical report, arxiv:1804.05116.

*QIP 2019, best student paper prize.* - Robust self-testing for linear constraint system games (Abstract)
with Andrea Coladangelo. Technical report, arxiv:1709.09267.

Presented at*QIP 2018.***(**Video) (Slides) - Separation of finite and infinite-dimensional quantum correlations, with infinite question or answer sets (Abstract) with Andrea Coladangelo. Technical report, arxiv:1708.06522.

Mathematical Notes |

- A 3-day course on ordinal arithmetic, taught at Canada/USA Mathcamp 2015.
- Slides for Wizards vs. Time Travel, Jan 6, 2017, Caltech Undergraduate Math Club.

Abstract:(Click to show)

Frequently Recommended Resources |

For thinking in general

- Thinking, Fast and Slow by Kahneman
- Rationality: AI to Zombies by Yudkowsky
- Algorithms to Live By by Christian and Griffiths

For doing theory

- An edX course on quantum cryptography taught by Professors Thomas Vidick and Stephanie Wehner.
- Lecture notes on quantum computation by John Preskill
- Scirate, the easiest way to catch important posts to arxiv/quant-ph
- Canada/USA Mathcamp, a summer program where grad students teach cool theorems to bright high school students. If you're in either group, do apply!
- Quantum Computing Stack Exchange. A community still in its infancy --- a good place to ask questions about the field, both general and specific.

Diversions |

- Ditch Day for Cranks, a fake Ditch Day stack (read: themed run-around and puzzle activities) based in part on parodying Professor Nets Katz' wonderful course Ma1a and associated text Calculus for Cranks.