This is the bimonthly 'What are you working On?' thread. Previous threads are here. So here's the question:

What are you working on? 

Here are some guidelines:

  • Focus on projects that you have recently made progress on, not projects that you're thinking about doing but haven't started.
  • Why this project and not others? Mention reasons why you're doing the project and/or why others should contribute to your project (if applicable).
  • Talk about your goals for the project.
  • Any kind of project is fair game: personal improvement, research project, art project, whatever.
  • Link to your work if it's linkable.
New Comment
59 comments, sorted by Click to highlight new comments since:

Working on "The Open Problem of Intelligence Explosion Microeconomics" and "The Open Problem of Lob-Tiling AI" (these are both tentative titles).

Is the former with Robin Hanson?

Nope, Robin Hanson did accelerating change microeconomics that didn't address the question of returns on investment in improved cognition, just the ability to convert capital in the form of CPUs into human-equivalent skilled labor.

But surely he would be interested in the latter...

By latter you mean the Lob-Tiling paper / article?

Oh, I meant he would surely be interested in a paper that addressed the question of "accelerating change microeconomics that did address the question of returns on investment in improved cognition".

Well, if no-one else is going to dive in ...

I have been feeling bored and tetchy for a while now. So for one project I took on the job of volunteer sysadmin at the well-known wretched hive of scum and villainy RationalWiki. It's been quite fun robustifying a small LAMP server overwhelmed by its Alexa 12,000 rating. MediaWiki is a lardy beast, but a coupla Squids in front and it's pretty good; ticking along nicely. Of course, now we have to stop the Squids from falling over too ...

This obviously didn't soak up enough energy. So the other thing I've been doing is the official unofficial RationalWiki blog - a skeptical blog. It's not official yet, but aspires to be; so it's stuff about RW itself mixed with more general skeptical interest. So far it's getting about 150 hits/day, which is not terrible starting from zero. I've also got the keys to @rationalwiki, for further social media delight.

So far, doing the blog feels like NaNoWriMo, or thirty things in thirty days. Thankfully I've actually talked other editors into writing a bit.

LW aims to raise the sanity waterline. But to extend the metaphor, sanity's waterways include quite a few alligator-laden swamps, and toxic waste spills in need of cleanup. I'd like RW to be good at that.

Out of curiosity, is there any sane way to have a debate with RW's top people about whether it's actually productive for them to go around systematically heaping scorn on anything that seems-to-them like an easy target (e.g. homeopathy and cryonics), while also scoffing at the nerds who try to talk about it using math? Or is there no probable productive outcome of even trying to have that conversation?

Basically not, because, like Wikipedia, it is literally true that no-one actually runs it and there are no top people. There is no-one who can tell trolls or (worse) the reasoning-incompetent to just fuck off, much as it would be nice sometimes, and some long-standing RW users stand firm against it on principle. (There's a board, and some of them are having to be beaten about the head with Section 230 - they were elected as charity trustees, not as forum wizards.)

It's a mob of annoying skeptics, and the qualification to be a skeptic is only slightly narrower than the qualification to be an atheist. So the bottom half of RW has lots of shit from lots of arseholes. I would like the blog to project a much better image by being a lot better, in the hope of inspiring people to better stuff. e.g. I think the homeopathy in India post is actually important.

I actually liked all the snark on Rational Wiki's LessWrong page... RationalWiki pretty much defines itself as a wiki for making fun of things, after all...

Hey, is there any chance that you could add an automatic site thing like "This page has been heavily (%number changes) edited by a low (%number) number of people for the last %number months. It may not represent an unbiased spread of opinion." Looking at you, LW site on RW ..

You know, maybe I would like this idea more if the way you put it, in the context you put it, didn't give me the vibe of "feature implementation with an agenda". Perhaps there are better considerations for choosing new site features than the light in which it presents the LW article?... I mean, yeah, sure, if considered out of context, your proposal looks like a reasonable attempt to increase accuracy -- but if accuracy is really what you set out to maximize, and not LW reputation, then you should agree with the other side of the coin: if the page has been heavily edited by a high number of people, there should be a message telling people to be more confident than average (that is, the average confidence about RW articles) about its reliability. Now imagine a world where this would hold true for the LW article. If that makes you ever so slightly less inclined to agree with this other implication of the policy you just proposed, then perhaps it's not really increased accuracy that you want.

On a related note: Seriously, people. Someone, somewhere, doesn't have an all-positive or even mostly-positive idea about us as a group. Which is fine. If you really care so much about reputation, consider the contents of the criticism rather than the fact that there is criticism at all.

You know, maybe I would like this idea more if the way you put it, in the context you put it, didn't give me the vibe of "feature implementation with an agenda".

It's true! I do not deny it.

if the page has been heavily edited by a high number of people, there should be a message telling people to be more confident than average (that is, the average confidence about RW articles) about its reliability. Now imagine a world where this would hold true for the LW article

On one hand, I think in that world I would not come up with this idea. On the other hand, I don't think in that world the LW page on RW would be such that I would expect to have this agenda.

On a related note: Seriously, people. Someone, somewhere, doesn't have an all-positive or even mostly-positive idea about us as a group. Which is fine.

I just don't like that the page looks as if it represents RW consensus while it's basically edited by one or two people. I don't know how common this is on RW in general.

Pretty common. (Also common in the long tail of Wikipedia, I think.) The current version is mostly AD's recent rewrite attempting to make it calmer.

I haven't heard of such a thing, and in any case most RW articles that are actually good fit that statistical description. There's no de jure ownership of articles, but in practice editors have their favourites and others often don't bother. (I suspect the same largely applies to Wikipedia, though that's just my human simulator talking, not numbers.)

I think (and IIRC people have done a few studies on that) that in the typical good Wikipedia article nearly all the content was contributed to by a couple editors, but there will be dozens of users who've made stylistic changes.

Working on an iOS (and hopefully Android) game based on the dual n-back task. I'm about a week and a half in and I'm still evaluating frameworks, but over the last 3 days I've built a working proof of concept in the Sparrow Framework which is iOS only. Over the next few days I'm going to port the proof of concept to cocos2d-x and make a final determination on technology by the end of the week.

PM me if you're interested in beta testing. I expect to have a playable and hopefully fun prototype (about 75% feature-complete) in about 3 weeks.

For some context, this is not an idle side project: I recently quit my job at a large game company to pursue this project, and I have several years' experience in the industry. I will be working solo and full-time on the game at least for the next 3-4 months, which is about how long I expect development to take to the point of completion. I plan on outsourcing the music — if you have any suggestions in this area, suggest away! I am well aware of the planning fallacy and Hofstadter's law, so feel free to take my estimates with a grain of salt, but please note that I do have experience estimating the cost of software projects and I think I am reasonably well-calibrated in this area.

Also, this is not just another Jaeggi or Brain Workshop clone; I am aiming for super-tight gameplay and mainstream appeal.

I currently working for a large game company, and while I have no plans of leaving, I feel as if at some point I would like to branch off and work on a more self-motivated project.

The cost analysis of this isn't easy for me. My position in the game industry is somewhat atypical; I joined one of the most promising teams in the world directly out of college, and have since done good work on our recently released project. From within the studio, advice on where to direct my career seems pretty risk adverse. We have high paying, extremely stable jobs which aren't guaranteed to be reclaimable if we exit, and with rewards becoming more lucrative the longer we stay. Two questions for you then:

Why did you recently quit your job? Is you main motivation along the lines of expectations of future earnings, a satisfaction in the intellectual worth of projects you choose to work on, or satisfaction in the worth of your product to others?

Where would you have ranked yourself in terms of your success in the large-scale game industry? I have a hard time comparing my situation to others, because as mentioned I feel as though my case is atypical and thus with potential for more higher payoffs if I choose rightly with my future. Having worked only a year and the half in the industry, I will be giving a talk at GDC this year about work done on our last AAA console title, which seems on the surface like something to make that case, although I recognize the major risk of self-aggrandizement here.

What city do you live in, if I may ask?

Your situation does seem unusual, so you'll have to do your own research and analysis. Speaking for myself, I am not strongly motivated by money beyond the amount required to support my standard of living, plus a healthy amount of buffer. Instead I have been motivated more and more by novelty and the opportunity to learn: by, in a word, curiosity (which is one of the 12 virtues of rationality, dontcha know ;) ). It also helps to be working on something I care about, have control over, etc. And I see this as a stepping stone towards projects that might have a larger positive impact. If you are attempting to maximize your income, you may want to stay with your current gig, but your terminal values are yours to decide.

I live in San Francisco. You?

Seattle, although I will of course be in San Francisco for GDC in late March.

I'm not yet sure what my goals are, although savings is a big one. If I knew I was going to be in the game, and even software industry forever, it might not be, but I've always wondered if I might not want to have the money to allow me to feel secure in later supporting myself if I chose something new that payed very little.

Also, I recently started a blog, which is dual-purpose. The secondary purpose is to document my progress in building the game, which I haven't done yet (there's a mobile game frameworks post coming up).

The primary purpose is to give me a place (and hopefully a reward structure) for writing about ideas that I think are worth taking seriously, including LW-related concepts. Part of the reason for this is that my peers, while for the most part smart and capable, are generally unwilling to read and engage with the articles on this site; the inferential distance is too large. So I am trying to write about these ideas in a way that can be understood by an intelligent person who hasn't read the sequences. The first and so far only post is in this vein.

I am finding that blogging is rather more difficult than writing software, and, for that reason and others, I am less certain of the success of the blog than of the game.



The paper has been finished, and has been submitted to a journal. Now I'm working on four new projects, more or less concurrently:

  1. An extension of the previous result a more complicated (and hopefully more realistic) model.
  2. A technical result in a related area, for a different kind of tomography.
  3. A completely unrelated theory involving scattering on orbifolds.
  4. Some numerical results for the previous article and (eventually) number one.

The boyfriend and I are still together, and I'm planning on moving in with him in July, if all goes well.

My main project right now is applying everything I learned at the CFAR workshop. The most visible thing so far is that I have a reasonably functional GTD system now (RTM + blocking time to go through RTM by tag with Google Calendar + Boomerang for very time-specific things). Using this, I have:

  • Started going to the gym (also used a social commitment mechanism for this)
  • Started an OkCupid account
  • Contacted 80,000 Hours
  • Processed the CFAR workbook into Workflowy
  • Started on my friendly AI reading list

I also have RTM tasks that will get me started on applying the rest of what I learned.


Previously on Battlestar Galactica. I found out that everyone else in the galaxy cluster biz was doing linear regression all wrong, so I had to write up how to do it correctly and then apply my method to some new data. I wrote the paper and it'll be submitted soon. In other news, JAGS is really neat.

Now I'm applying for summer research gigs.


Since December, I've been persuing a "remedial computer science education", for the sake of both well-roundedness and employability. My background is in the purest of pure math (Ph. Dropout from a well-ranked program), so I feel I can move fairly quickly here, though the territory is new.

My biggest milestone to date has been solving the first 100 Project Euler problems in Python (no omissions!). I had had a bit of Python experience before, and I picked 100 as the smallest number that sounded impressive (to me).

Second biggest milestone: following a course outline, I wrote an interpreter for a very limited subset of Scheme/Racket. This really helped de-mystify programming languages for me. (Although rather than learn OCaML like the course wanted, I just hacked it together in Python so that I could move on to a new project sooner.)

In the same vein, I'm currently reading and working through SICP, still using Racket. I'm in Chapter 3 of 5, though I'm often peeking ahead to Chapter 4 because it looks pretty exciting.

Of course, I won't be a true LISP wizard without understanding macros, so the next (or concurrent) project is to go through the relevant Racket Docs tutorial.

I have some other likely future projects in mind, though I'm actually trying not to plan too far ahead lest it all appear more daunting.

  • Forcing myself through some C, to build character. This was explicitly recommended by a software engineer friend as a more "useful" way to spend my time than learning to LISP.
  • An algorithms course, possibly using this book

It turns out that I'm not very good at doing personal projects when I have a job. In order of most-recently-done-anything-on:

  • I'm writing up a solution to the shooting room paradox. I've been doing this for about a month. Currently in "proofreading and latexing" stage.

  • I wrote a debugging tool which I've found useful a couple of times at work, but haven't been successful at getting feedback on. Realistically, there's a good chance that I won't touch it again until I need it to do something that it can't yet.

  • This has been pushed aside by the shooting room, but I've been working through PT:TLOS and typing up solutions to the exercises.

The shooting room paradox is a neat one.

Later your mother learns that you entered the room. How worried should she be? Extremely: About 90 percent of the people who played this game were shot.

... 90% once the game is over, but 0% if it's not yet, so it all depends on what other information is available to her, right?

That's what I thought, but no. Make the game last at most an hour with exponentially decreasing round lengths. Consider what happens if your mother watches you enter the room, then falls asleep for an hour.

Make the game last at most an hour with exponentially decreasing round lengths.

Assuming you mean that the game is guaranteed to be completed within one hour, it is equivalent to a potentially infinitely long game with fixed round length, and we are back to my point.

When she wakes up, she knows the game has ended. Before she goes to sleep, she knows she will know this when she wakes up. So how can her probability estimate change between the two events?


When she goes to sleep, doesn't she not know whether you will play the game? After all, the number of 'players' depends on how many rounds it will take before double sixes are rolled so the only guaranteed way for you to be picked is if you were first in line to be picked.

I'm supposing that she goes to sleep right after you've been chosen to play (i.e. you've been called into the room, the dice are about to be rolled).

If she knows that the game will have ended, she knows that the odds will be grim. Not sure what two events you are talking about.

I think you are claiming that

  • When she goes to sleep, she believes you have a 1/36 chance of dying.

  • When she wakes up, she believes you have a >90% chance of being dead.

But the only thing she learns on waking up is that the game has finished, and when she went to sleep she knew that she would learn this. This violates the maxim "if you know your destination, you are already there".

I think you are claiming that When she goes to sleep, she believes you have a 1/36 chance of dying.

Err... No, my point was quite the opposite.

Oh, I see. In that case, when you enter the room, why is her probability estimate different from yours? (Or if it's not, why is yours >90%?)

Different levels of knowledge result in different probabilities... You have more information than she does. Your calculation computes the prior probability of you dying during one round. Her calculation computes the posterior probability of you dying during the whole game, given that you played and that the game has ended. The paradox only arises if one treats probability as an objective thing.


When you've just entered the room, what knowledge does one of you have that the other doesn't; or why are you computing the probabilities of different events?

She has (will have) the knowledge that the game has ended, you don't. If you could know that the game ends in this round, your probability of dying would be 100%.

She knows the game will end / will know the game has ended. She doesn't know which round it will end / ended in. You also know the game will end, and not in which round; and you know what she will know upon waking up.

Looks like we are talking past each other, so the only way to continue is to show the calculation:


  1. P(you die in this round| you play this round) = 1/36

your mom's:

  1. P(you die|game is over) = P(you play in the last ever round) = 9/10

You can express P2 by summing P1 over multiple rounds, weighted by the odds of the round being last and by the odds of you playing in it. But the important point that P1 and P2 are probabilities of different events.

And with that I am disengaging.

Maia and I have been working on a chore market for our house, and we just started using it. People bid how much they want to be paid (in points) to do a chore, and when the auction closes everyone is taxed to pay for it. If you go into debt, you're forced to bid on chores. This is the closest thing to a "major software project" I've ever done, so it's pretty nice to see my baby actually sort of working.

It's on Github if you're interested, but it's pretty buggy at the moment and not well documented, so user beware.

Q. [any problem whatsoever]

It's worth a try, and if it works then excellent :-) But I suspect the sharehouse problem is not a problem of mechanism.

IME, all I've found to work is every person personally considering household chores their own problem. (Approximately: If you all feel you're doing 1/N + 20% of the work, it's probably about even.) If even one doesn't, things get crappy and someone suggests a mechanism. Bogged-down household resentment ensues and it all goes a bit Strategy of Conflict.

(Some of the housemates I've had, the friendly AI would choose "nuke from orbit" as the only reasonable answer.)

It strikes me as quite likely that I am speaking through personal post-sharehouse stress disorder. Further reading: John Birmingham. 1 2. The standard texts on the subject. Australia in the '80s and '90s, when the system really did love you and want to be your friend.


The system operates on a sort of baseline of trust anyway, so it's not too likely it would work in a situation where some roommates are unwilling to pull their weight. The goal is more to give people a better idea of how much work they're actually doing.

Also, its value to us in chore-distribution-optimization has so far been far less than its value as a fun project and an interesting idea to discuss (and I anticipate the latter being more valuable in the long run, too).

I am (being new here) working through the sequences. I do this because I've immediately recognized upon reading many of the responses here that I lack the internal machinery, and certainly the language, to even foster discussion of what I do and believe, and why. The goal would be to get through all of the major and minor sequences, which seems like quite the low bar, except that I work a reasonable amount of hours throughout the week, and suffer a good deal of lethargy over not spending enough mental effort on reading new things already. On that note, the major portion of this project is recreating the ability to focus on material for longer periods of time without feeling like my brain needs to shut off after using some of its energy for work activities. While work is, intellectually, hard work it is of a tried and old sort for me that allows for lethargy without a decrease in quality, a feature I would expect of most jobs.

Have you tried reading through Methods of Rationality first? That's more fun and covers some of the same material.

Admission: I haven't read Harry Potter, but I'm told it's not a major prerequisite.

Well, you won't understand some of the jokes if you don't have some familiarity with the original series, but that's nothing a few trips to Harry Potter wikia can't fix.

  • Coursera's Data Analysis (basically applied R); it's going well so far.
  • my second (blinded, standardized doses upon awakening) potassium self-experiment is nearing completion; my subjective estimate is that there is still some damage to my sleep. Another longer-term self-experiment should also finish in February, but I'm not sure whether I will discuss it publicly.
  • working on another big expansion of my hafu database; I discovered that it's actually pretty trivial to scrape the HTML into an R data frame and manipulate, so I can finally see what the data looks like, but I also noticed that some of the data was messy or wrong and that I had omitted hair & eye color. This motivated me to clean it up and spend 2 days adding that data, and soon I'll start looking for entries from my last source of leads. Then I can finally write up an analysis and have something to show for all my time.

I'm working on an algorithmic music system. The (likely unachievable) goal is to make a generative clone of my own creative process. It's a kind of catharsis, bringing every unspeakable intuition bubbling up to the surface to be translated into a hierarchy of computational processes. Lots of trial and error, red herrings, and accidental success. My short-term goal is to be able to co-perform with it live in an entirely improvised manner, maybe by the end of 2013.

The current focus is 'groove', and it's been a nightmare trying to find a common thread between dissimilar patterns that sound good. If anyone know of any good papers on the subject, please share! I'm just going to throw some boltzmann machines at the problem before moving on to something else.

Other than that, just some Coursera classes, finalizing a DIY 16 bit MIDI->CV DAC, and transcribing a really good flamenco record.

Studying for the Level 2 (of 3) Chartered Financial Analyst (CFA) exam in early June.

In addition to hopefully getting a valuable credential, I'm learning about finance stuff.

Still in the planning stages regarding an article about the fair value of internal non-traded liabilities such as bank deposit liabilities. I'm hoping to show that a temporal dimension can be added to the concept of (wealth) utility in a consistent way.

This is a follow up to the last time I posted in a WAYWO thread.

A little over a year ago, I started working on a code searching tool in my spare time. It's been more successful than I ever thought it would be. The GitHub repo has more watchers than Ack, the project I set out to imitate. I learned a lot about optimizing, profiling, benchmarking, and using pthreads.

It's also had a nice side-benefit: random people online recognize me.

I frequently have cause to grep -r our entire codebase, so I think I'll be having a play with this. Thank you!

Edit: And you've helped me impress my boss. Well done :-D (We use Ubuntu 10.04 VMs as servers, so the Lucid deb was just the thing.) HEY EVERYONE: If you ever find yourself having to grep -r a codebase, just use ag instead.

Feature request: output in the same format as grep -r (to feed to other scripts). Edit: and lo, it does this automatically when its output is a pipe. Win!

Playing around with search-space heuristics for more efficiently approximating S-induction.

Which actually sounds a lot more impressive than the actual thing itself, which mostly consists of reading wikipedia articles on information theory, then writing Python code that writes brainfuck (decent universal language).

EDIT: Also writing a novel, which is languishing at about the 20,000 word mark, and developing an indie videogame parody of Pokemon. Engine is basically done, getting started on content creation.


I wrote a task manager (to-do manager) for myself on January 1, and have been growing it since then. The user interface is inspired by Taskwarrior, but I use an sqllite backend, and therefore it's 300 lines of Python instead of 30k of C++. The small size allows me to be flexible in testing various ideas I have around task management - a new feature is usually just one or two SQL queries away.

My most promising exploration has been to not accept any tasks to be older than 2 weeks. If I haven't managed to do it by then, there's something wrong - it's ill formulated, or there's some unsatisfied dependency that I'm not clear about, or I don't really intend to do it. I've developed a check list to go through to collect some stats about why tasks get to that point, with the intent to recognize them earlier.

One of my major failure modes seems to be to totally overestimate my capacity to deal with people the longer I think into the future - I'm totally sure I'll be able to confront my landlord in a week, but not today. Much worse than the usual discounting happening all the time.

Nice! I'd be curious to see some of this.

I was just thinking of exploring options in todo-list-management software, as there are a few features and algorithms I'd really like to see in such a program that I think might be really useful to help act and manage the todo list(s) more easily (would need testing, I'm not certain they would be, just a strong hypothesis).

Still doing well at GJP - my major recent achievement is that I'm no longer just at the top of my team's scoreboard; my score is better than the median score of any other team.

I'm talking about this at various software development related conferences, because it strikes me that the mere fact that it's possible to improve markedly at forecasting is well worth sharing. I start by encouraging folks to experience how badly calibrated we are without special training. A basic estimation exercise gets the point across well enough. I then move on to various ways practice makes perfect: PredictionBook, GJP, personal experiments, etc.

(I'm aware of Eliezer's reservations about mentioning calibration and overconfidence without proper warning. My experience so far has been that the experiential approach, where people first experience how far off the mark their feeling of "90% certain" is, doesn't seem to turn them into "sophisticated arguers". This could be selection bias - so far I've offered this talk in small gatherings of relatively sane people I already know. I'll be on the lookout for warning signs in larger conferences.)


Promoting the works of Dora Marsden (1882 – 1960). Egoist, freewoman, suffragette, author. Two Kindle books and one weekly blog of quotes. These are the first published compilations of her essays. Read about her from me or from elsewhere, but read about her.

Other resources for Dora:

This is what I was working on last time. Next time, I assure you, something different and equally grand.

  • Preparing for my oral exams to advance to candidacy

  • Messing around in GameMaker to prototype lots of different ideas (I currently have a prototype of a game where you use matrices to defend yourself against vectors)

  • I signed up for CodeAcademy with the intent of starting to dive in next week because my two usual free nights I had to work late this week. I would appreciate some external pressure to actually diving in by the time Monday rolls around.

  • Running a tabletop pokemon game which is a lot more serious that previous tabletop games I've run. I'm looking forward to the increased philosophical influence over my players. >:3

  • Looking for more ways to fill my time that, in the future, I will wish that I had done at this point.