Painful writing

Writing is hard for me. It’s always been hard for me. In high school I agonized over 5 paragraph essays. In college I dreaded research papers and fretted over them. Whatever I wrote was going to be bad and would never feel acceptable to me.

I force myself to write anyway because I think I have useful ideas that are good to share with the world and because writing things down helps me clarify my thinking. And there’s something elegant about putting my thoughts down in permanent form. If I have an idea and write it down, it becomes undeniable that this idea was real and I had it.


When I sit down to write a blog post, generating writing is often pretty easy. But turning jumbled writing into polished writing and verifying that the writing is okay, that it conveys the meaning that I want it to convey and isn’t objectionable in some way, is usually painful.

Reading back over my writing, and especially looking over writing that I haven’t just written fills me with dread. It invokes the same feeling as listening to a recording of yourself talking. Rereading older writing of my own feels terrible because I see flaws in it, but don’t really understand them and there’s nothing that will prevent me from making similar mistakes in the future. Because I don’t understand the flaws, it’s hard to say how bad they are.

It also feels painful to imagine different ways that my writing could be. I can look at a paragraph and try to make some improvements to it and that often feels fine. But writing a different version of a sentence I already have and comparing them is painful. I find myself thinking ‘no I couldn’t possibly come up with a different way of saying that’. If I do come up with sentences to compare, actually doing it feels slow and difficult since its hard to see what makes one better than the other.

I noticed recently that as I read my own writing, I worry that someone will come along and point out that it’s terrible, and obviously so, and that I will be embarrassed. That they will point out that all my sentences are awkward and horribly confusing. Or that my writing has a quality distinct from virtually all other writing marking it as childish. No matter how much I look over my writing or have others review it, I can never seem to eliminate this fear.

What’s going on?

I don’t know what’s going on here, but I suspect it has to do with not being able to see what my writing is mechanically accomplishing or not having a felt model (in the Focusing sense) of how writing works. That a paragraph is introducing an idea, or this next sentence is providing some important context, or that you need all three of these examples here because with only two it would be ambiguous.

Because I can’t see these mechanical properties of my writing without explicitly looking for them, any flaw that becomes visible becomes the only visible thing. It’s hard to keep the fact that the writing was accomplishing something before in mind. My internal model seems to be that my writing can be valuable as long as it doesn’t contain any flaws. But any flaws render it valueless, and all flaws are secret flaws only visible by other people with the special kind of seeing to understand writing.

I also I have a sense that I’m missing some kind of basic concepts in writing. That there are a few basic facts about how people read writing that shape how writing works that I don’t have. This explains why I’m afraid my writing has hidden serious flaws; if there are basic effects that I don’t understand, my writing may seriously harmed by a basic effect, and I won’t see it.

Putting these together, an obvious hypothesis is that I’m missing a felt model of how writing works because I’m missing some basic concepts about how people read writing that strongly affect how writing should be and generate some basic facts about how writing should be.

How to improve things?

The theory that I’m missing a felt model because I’m missing some basic concepts suggests I should figure out those basic concepts about how people read. But how do I do that?

I’ve spent some time with my brother Erik helping me talk out loud about what different parts of my writing are accomplishing more or less sentence by sentence. That was interesting, and I think it gave me a bit more of a felt model in some places, for example, that the purpose of transition sentences is to set context especially about the intent of a section, but I don’t think I could develop a full felt model this way.

Another idea for developing a felt model of writing is to have someone else explain to me how my writing is working on a low level and questioning them about why it doesn’t work a different way. I can imagine stumbling across new fundamental principles this way. This feels intuitively attractive, but it feels like getting the right thing would be tricky.

I could instead work develop a strong felt model of how I read, so that I can derive a felt model of writing. But how would I do this? The basic approach of paying attention to my internal experience while I read feels like it would be fruitless. I expect the way I read will be opaque to me because the important details will too low level and I don’t have good ways to probe to what’s going on.

On the level of figuring out why writing is difficult, I was able to notice the things in this post by paying attention to what writing actions felt painful or aversive as I contemplated doing them, and keeping my internal peripheral vision looking for fears that flit by for a split second. Hopefully continued attention on the details of writing and reading will reveal how to develop a better felt model of writing.

A model of anger

Epistemic status: I feel confident in this. I find it useful. But its mostly based on introspection.

First a story.

Once, I was having a conversation with a friend about not wanting to draw diagrams to help me strategize because it didn’t feel like it helped. As we talked, I fretted to myself about sharing a vulnerable point: that when I write things down or make diagrams for myself, I feel a strong urge towards making my diagram as formal and complete as I can. I often feel bad about not being able to make my diagram more formal, and give up on making it. I knew abstractly that this urge was harmful, and in the conversation it was obvious I was wasting time on it, so sharing it would open me up to ridicule and judgement. But I felt just barely safe enough, so I shared it anyway.

The first thing my friend said was an exasperated “John, you’re going to have to stop doing things like that.” Which hurt a lot for a split second.

But I had been training to respond with anger sometimes, and I was able to respond by shoving them in the shoulder and saying “I fucking know that. Can you listen to me?”. My friend apologized, saying they had been quick jump to conclusions, and we continued the conversation productively.

This whole exchange took probably 30 seconds.

My expression of anger let me protect myself from my friend attributing an unsophisticated perspective to me and ending up with a poor opinion of me.

The most interesting thing is that I think that my ability to be angry with the made it more possible for me to share this vulnerable point in the first place. If my friend threatened to think badly of me, I had a way to respond. I would not have shared it without that possibility.

What is anger for?

I think this question and several like it are extremely useful. What information does it carry? What is it trying to get me to do? And given that, what is the right way to respond to it?

My current theory is that you feel angry when one part of you judges that one of your allies is undervaluing your welfare in some way and that it’s correctable.

This makes sense of who you feel anger towards. You only feel anger towards people who you think could and should take your welfare into account. You’re much more likely to feel angry towards your friends than strangers. And when you do feel angry towards strangers it’s usually accompanied by a feeling that they’re ‘supposed’ to take your welfare into account. You won’t feel anger towards invading aliens (though maybe aggression), but you will feel anger towards your roommate leaving crap on the floor.

On this theory, the proper response to anger is to show the person you’re angry at that you see that they’re undervaluing your welfare and that you’re willing to punish them to make them stop. This is good for the relationship; the other person will be less tempted to undervalue your welfare if they know you will often catch them.

I have had an extremely difficult time expressing anger. It seemed to be cruel to be angry at people. But now I think it is useful and healthy, especially in small, short bursts (the most information dense).

Anger is there for a reason: it’s telling you about the world. And understanding what anger is trying to communicate can help you respond better to it.

Edit (09/25/2016): All emotions are there for a reason. See also this model of shame.

Many thanks to Katja Grace. I developed this theory through discussions with her and could not have done it without her.

Short term prioritization

It seems like short term prioritization is a really useful skill. Its something I can do reasonably well, but I feel like there’s something I’m missing.

My current habits work like this: I often write down a list of things my brain generates as important to do. Things that my brain is like “you IMG_20160106_133559should do this.” Often my brain has been pinging me about several of
these for the last half hour. For example, today this looks like:

  • Theorize about x-risk
  • Plan day
  • Plot about coffee?
  • theorize about epistemics
  • plot about fashion workshop
  • w2p
  • pymc3 thing
  • planning request

Most of these don’t mean to me what they seem like they mean. They’re just the first word handles I found. For example, “Plan day” really means something like “make sure I leave enough time to get to a meeting in SF and check that I know where I’m going”. “theorize about epistemics” means something more like “think about why I feel uncomfortable talking to people in a way that seems linked with not getting better at how I pay attention to evidence”.

Writing down these word handles lets me quickly flit between different options more easily and thus clears some working memory for thinking about the choice as a whole. My next step is to go over the list and mentally note, circle or rewrite the ones that seem especially yummy or anxiety inducing.

Then I’ll select 2 or 3 from those as my next action items that I’m actually planning on doing and write them on a separate list. Having this short list makes me feel more decided and motivated to start them. I cross them off as I do these tasks and when I’m done I refill the short list with more items. Currently I’ll also write down items I do even if they weren’t on original list and then cross them off, but I suspect this may not be good.

While evaluating items, I’ll often pay attention to each item and try to evaluate how important it seems. I’ll often note if its urgent in some sense and whether I’ll have the mental energy to do the task. Sometimes I try to pay attention to how valuable an item is, but my methods for doing this seem pretty crude and based mostly on affect and maybe briefly visualizing what other things in my life the task affects.

Its this last area where I think I could improve the most. I need better heuristics or methods for evaluating the tasks values. Perhaps I could do fermi estimates for the value of different tasks, but this seems pretty difficult, especially for reflection-type tasks (which are often the most valuable), and I suspect I won’t do it. Maybe I could do very short free-writes about why the difficult ones are valuable? Or just writing down the main effects I think of?

I’m supremely curious how other people do short term prioritization like this.

PyMC3 Beta!

Samples drawn from a stochastic volatility model.

Samples drawn from a stochastic volatility model.

Probabilistic programming allows for flexible specification of Bayesian statistical models in code. PyMC3 is a new, open-source probabilistic programmer framework with an intuitive, readable and concise, yet powerful, syntax that is close to the natural notation statisticians use to describe models. It features next-generation fitting techniques, such as the No U-Turn Sampler, that allow fitting complex models with thousands of parameters without specialized knowledge of fitting algorithms.

Stochastic volatility model

Stochastic volatility model

PyMC3 has recently seen rapid development. With the addition of two new major features: automatic transforms and missing value imputation, PyMC3 has become ready for wider use. PyMC3 is now refined enough that adding features is easy, so we don’t expect adding features in the future will require drastic changes. It has also become user friendly enough for a broader audience. Automatic transformations mean NUTS and find_MAP work with less effort, and friendly error messages mean its easy to diagnose problems with your model.

Thus, Thomas, Chris and I are pleased to announce that PyMC3 is now in Beta.

Try it out!

To get started with PyMC3, I recommend the Tutorial.

If you have a question, we are quite responsive on Stack Overflow and Twitter (@johnsalvatier@fonnesbeck and @twiecki). If you have a bug report please post it to our issues list.

Digging into ACE: Priors

I and others are digging into Animal Charity Evaluators, diving deep on their research, to give them critical feedback and give other people an outside perspective. Before doing (most) research I’m recording my priors. My primary target is their fundamental research on leafleting.

I’ve already done a little bit of digging, so this isn’t totally unbiased.

ACE’s numbers on leafleting look unrealistically good, so I expect our investigations will push that number downwards. I expect they are not being pessimistic enough. I do expect them to be very open about their research.

I expect ACE to have been pretty thorough in most areas, but to have made some significant errors in some spots that we will find that make the leaflet effectiveness numbers go down. I worry that our modifications will make leafleting seem drastically less effective, but I guess it will just look somewhat less effective. I worry that we they will be reluctant to change things based on our feedback, but that’s not really based on anything.

Meetup Retrospective: Donation Decision Day 2014

At the end of 2014 (December 28th), the Seattle Effective Altruists group ran a hackday/workshop called Donation Decision Day for people to come and make their final donation decisions for the year. The idea was for people to come work out how much they wanted to donate and where they wanted to donate to. If they got stuck, they could talk with a partner to help troubleshoot their situation.

Overall, the Donation Decision Day was very successful, and will definitely be a yearly event. I think other groups might be able to copy it successfully. What follows are some thoughts and observations about the event. The Seattle group is unusual in that it’s mostly working programmers instead of students, so other groups holding similar events might look different.

Several people expressed some confusion about what kind of event it was going to be (event description at the bottom), so perhaps some people were scared off the event by a confusing description. Need a better name and description next year.

Since this was an especially action oriented event, I had hoped we would get people who had only been lightly involved or not involved at all to come, so the event would serve as an outreach event. But, that didn’t happen at all. 11 people showed up (about average), but they tended to be the most committed of the group, not the lightly interested. Next year, I would like to try specifically to get new people to come, as I think this is both a natural introductory meeting and a high impact activity. Not sure how to do that, though. We’ve discussed trying to run similar outreach events at our workplaces, but haven’t made concrete plans to try this.

A couple people were giving nothing or smallish donations, several people giving 5-20% and one person giving significantly more. I liked that there was a big spread of what people felt comfortable with. People lightly encouraged each other to give more during discussion of how much to give, but I think mostly avoided pressuring each other.

One person told me they gave dramatically more because of the meetup than they otherwise would. I think I probably gave about 10% more than I would have otherwise definitely was more efficient about it. Several people had productive 1-on-1 discussions about where their donations would go. We also coordinated with the Vancouver group to trade AMF and Give Directly donations, which allowed the Vancouver group to get the tax deductions on their donations and donate more.

The event description we used:

Lets get together and work on making our final donation decisions for the year!

Doesn’t matter if you’re giving $20,000 or $20.

Doesn’t matter if you’ve never done research on charities, or if you’ve read the primary literature on malaria net research.

Wherever you’re at, try to beat your personal best: give a little more this year and do a little better research this year.

Come and work on your personal donation decisions; with a partner or by yourself. Do research, run the numbers, and if you run into trouble, have a 1-on-1 discussion with someone about it.

Once you’ve made your decisions, send out your checks with our stamps!

Come starting at 11am to have a short discussion about donor advised funds etc.

People will be here all day, so feel free to come at any time.

Donations for 2014

In the sprit of normalizing talking about charitable giving (and to encourage myself to push myself a little harder), I’m going to do something uncomfortable and brag about my giving this last year.

Last year I gave 17.5% of my income to charity, up from 13.5% the year before. Which is a personal best.

65% to GiveWell unrestricted
15% to Against Malaria Foundation
10% to Center for Applied Rationality
05% to Machine Intelligence Research Institute
05% to Animal Charity Evaluators

This year, I want to plan my donations more in advance and spend more time thinking concretely about where my donations have the highest marginal impact, and to push myself a little harder.