Published on: 2020-07-01
Warning: A bit of a rant, mainly writing it for myself, but I feel it's appropriate to make it part of the blog. Read at your peril, I don't guarantee it's appealing or interesting to anyone, I do however guarantee that the writing is of much lower quality than usual.
Recently I've unpublished some articles on the grounds of them being utter garbage, not the kind of garbage that at the time I mistook for sound thought, but the kind that even back then I felt unsure about writing.
They mainly fall into the "tutorial" section, things which I am now certain were utter gibberish written for the sole purpose of social signaling alone, not as a way to add rigor to my own thoughts.
I notice these kinds of "articles" often and they annoy me to bits.
For example, I'd want to learn about machine learning by reading accessible yet insightful work like that present in some machinelearningmastery articles or in the fast.ai courses. However, that kind of work often gets buried in the social signaling game of "Look, I wrote a paper/article/tutorial/code-example about this hot topic in ML, that means I am an expert in this field and you should trust me and prioritize hiring me over other people for that slightly higher paid but ultimately shitty job".
Even worst, the line between genuinely presenting one's thoughts on the subject and social-signaling is often blurry and not helped by the fact people are hardwired not to notice when they signal.
1. The failure modes of information transmission
To put it more generally, when you want to share information there are 2 potential failure modes:
- Sharing by repeating a thing that you do not understand nor wish to understand because it serves as a good social signal of you knowing that thing.
- Sharing a thing that you understand so fundamentally that you don't remember how the world would look had that thing not been known. You can state the thing perfectly well, but are unable to comprehend the position of your interlocutor.
The second failure mode is desirable since you at least know that the thing being explained is "correct" in some sense of the word. You're left to do all the legwork of internalization and determining whether or not the information is actually valid or useful.
However, even this second failure mode is desirable only in a select few fields where information has close to objective value. It is desirable in learning programming or nursing but not in learning psychology or philosophy.
Even worst, someone that doesn't know "the thing" is likely unable to differentiate between the two failure modes, barring external social information about the one exposing it.
For example, if I know nothing about string theory (and I don't), I might listen to Alan Watts and Edward Witten explain their ideas around it.
I would fail to understand what either are trying to get at, but it would also (probably) not be obvious which failure mode their explanation falls under.
But I could look at Alan Watts' language, his opinions about other subjects I do "know" and his general social standing a demeanor and conclude "Hmh, looks like a spiritual quack to me".
Conversely, I could look at Edward Witten's language, his opinions about other subjects I do "know" and his general social standing and achievements and conclude "Hmh, looks like a respected physicist and one of the people that put the basis for this field.
But now assume it's not Witten and Alan Watts that are trying to explain string theory to me, instead it's Roger Penrose and Garrett Lisi.
Well, Lisi is a physicist that seems to... ahm, do some things around the subject, at least as far as criticizing it goes. But he's not an academic, he's independent, and most of the community seems to be against him.
Penrose on the other hand is a well-enough respected mathematician, but he's rather old nowadays and physics is not quite his field. I've heard him talk about "quantum effects and consciousness" and on my spiritual quack detector that puts you somewhere around "man dressed in orange, speaking very little and selling hashish in northern India".
So now who do I trust? How do I know if they are falling in failure mode 1 or 2?
But even assuming I know, by divine intervention I get a book and I am 100% certain this is the most complete and correct representation of string theory out there, I still don't know if I should read it.
Will string theory help me out? How am I supposed to integrate string theory in my rather lackluster understanding of physics? Should I just drop the "layman" model of string theory I have right now to better understand the correct scientific model or should I build on top of it? Do I even have the required background knowledge or mental capacity to understand this? Is string theory actually correct in the sense of allowing me to think meaningfully about the world and understand&modify it through its lens?
2. Ways out
The solution to these 2 failure modes seems to be one of two things:
- You can be an amazingly intelligent person with a lot of knowledge and an exceptional memory and theory of mind, that is not only able to know a thing, but also able to know how it must be to not know that thing.
Renowned teachers like Robert Spowlsky and Feynman seem to fall into this category.
The problem with this solution is that there's no a lot of these people, and even they might only be able to do some of the internalization/usefulness/validity related legwork for their students. Their perspective might not fall into failure mode 2 for some students, but might for others who have minds that function differently than the teacher's theory of mind and/or have inappropriate background knowledge.
One other problem with this solution is that it seems to only apply to "old" information, new important discoveries cannot be shared in this way, because by their very nature the people thinking about them haven't quite understood their full implications or internalized them.
The last problem is that this solution tends to lead to the centralization of good information. When thinking about why some groups of people seem much less informed than others (e.g. think Rust Belt vs Northern West coast), it might be that this is rooted in immense genetic potential and/or cultural differences.
More likely, there might be a small difference in mindset between these two demographics, but a runaway improvement effect starting from a few good teachers has resulted in thousands of good teachers in one of the groups but not the other. It is not that one demographic is fundamentally harder to teach than the other, it just so happens that good teachers ended up having a theory of mind well suited to a demographic and this effect snowballed.
- You can try sharing things are you are learning them. I think there are a lot of good blogs and podcasts doing this, examples that come to mind are Shtetl-Optimized, Slate Star Codex, and Ribbon Farm.
This seems like the kind of thing especially conducive to very simple subjects where learning consists of just integrating a few basic insights. For complex subjects, where learning anything might involve years of head-against-the-wall beating, I haven't been able to spot examples of this approach working... Fastai comes to mind as the closest I can think of, but it's a very flawed project and Jeremy Howards is literally one of the smartest people currently living (so this kind of falls into the previous case).
The person learning the "new thing" matters a lot, when one of my friends tries to figure out how to lose 30lb any learning experience coming out of it is probably fairly pointless.
When Andrej Karpathy tries to lose 30lb, he generates an almost unmatched introduction to metabolic and molecular biology as a side-effect of doing so.
But I am not a wise teacher.
Nor am I Andrej Karpathy or Scott Alexander.
So the question remains as to whether or not I can effectively share some meaningful information efficiently using this "sharing as I myself am understanding" approach on my blog.
I don't know the answer to this question, but I can at least generate a cohesive purpose for this blog, something I couldn't have done a year ago.
I am trying to write down certain personal learning experiences to force myself to bring more rigor to them in the process. Hopefully seeing this thought-process laid out might also help certain people who are in a similar place internalize and critique these ideas.
Going even further, this style of explaining things might be mutually beneficial, since people reading the explanation might conclude it is wrong and voice their concerns. In an early stage of learning, this might help me with internalizing a slightly different idea or just discarding it, this is usually much harder once a thing gets entranced in your mind.
3. But not really
That being said, this is not an entirely satisfying answer, because most of the articles I write are not that... at least not fully so.
Named Distributions as Artifacts is a collection of ideas that have been fundamental to my rejection of many fields of studies in the past. And the article only delves into the surface level of the issue. I certainly didn't "learn" anything new by writing it, nor did I even bring much more rigor than before to my thoughts around the subject.
It's the kind of article that I write mainly to see if an insane idea that I hold has any appeal to the external world. I determine this in part by just reading the thing over and trying to put myself in the head of an interlocutor. And partially by looking at comments on LW, lobste.rs, HN, and the various subreddits where my articles sometimes get posted or where I share them myself.... to that, I can add the rare emails and direct comments I get.
If everyone thinks the idea is dumb, senseless, or outright against social norms, I try to rephrase it or even start to consider whether or not it may be incorrect (even if a direct refutation isn't provided). After all, there's some worth in holding ideas that nobody understands or agrees with, but they can often be dangerous if your brain is overloaded with them.
A bit of madness is useful, a lot of madness is harmful, even in the very few cases where history might agree with said madness 200 years since.
Many people seem to us like islands of sanity in past times, but usually, they were awarded with death or ostracization for their views and weren't able to help their fellow men. The ones who made a difference were the ones that were reasonable enough to be respectable with just one or two crazy ideas they remained passionate about.
Even then, I think there's a breed of articles that I'm not covering here. Like the article, I wrote about COVID-19 in mid-March, that one is just there so I have some public record of my prediction.
Or, I can think of it as a reminder of the things I got right. It's there for me to brag about figuring out the close-to-correct approach back when most people were proposing and applying an outright insane solution.
Or, I can think of it as a reminder of the thing I got wrong. I forgot to even touch on the subject of mask-wearing and I was convinced (granted, much like literally everybody else) that it was a respiratory disease.
Then there's the occasional story I write and those also don't fit any mold.
That being said, I think 90% of articles do share the "explaining my ideas and thoughts in order to help my learning process, get counterarguments, give them some rigor and figure out if any of them are outright batshit crazy or plain wrong and thus I should stop holding them".
Which brings me to the thing that is missing most from my Blog, comments. I've consciously disabled comments for the first year of this new blog.
My reasoning was that comments on the websites where I share my article are sufficient and I didn't want to deal with the implementation and all the potential spam.
Recently I've come to realize how central the role of reader opinions is for this blog to fulfill its purpose. So I think relying on 3rd party websites for this is a mistake.
To this end, I've added a very minimalist comment section about 2 months ago (using schnack) and now I've extended it to one that supports anonymous comments, more auth methods, email/rss subscription and doesn't require approval to post a comment (remark42).
I should note, both my current and previous comment systems are open source software that I run on-premise. I hold to my conviction of not tracking my users or offering up their data to malicious 3rd parties like disqus.
However, my experience with the previous comment section was that it was pretty dead. It had maybe 16 comments tops over the close to 2 months it was up, which is rather surprising considering how many requests for my index.html I get every day.
Granted, this may be because the comment section wasn't all obvious and the login options were limited (no email login and anonymous commenting).
At the same time, it might be because my blog is not focused enough on a single subject to build anything like "an audience". I have a fair amount of subscribers at this point, but it's unclear to me how many of them are actually actively reading this blog and how many of them just didn't realize they can reply to my subscriber emails with "unsubscribe".
One amazing thing I realized is that the number of views I get for a given article is fairly large for an internet blog (on average maybe ~9,000 per article for the last 6 I wrote, with the exception of one that went viral on HN and got over 100,000), but I suspect that most people don't finish the articles and/or aren't interested enough to leave a comment.
So at the end of the day, it seems like what I ought to do here is give the blog a bit more direction, that way I might hit three rabbits with one stone.
I'd get to delve deeper into subjects that interest me since I am limiting my options and thus I am forced to aim for depth.
I'd be able to explain things further away from the surface level since I can do it throughout multiple articles and assume common knowledge of the past articles.
I'd be able to hopefully get a smaller but more engaged readership from which wants to give me useful insights.
The main issue with this is that... well, it seems boring and limiting.
Maybe there is a middle ground where I release surface-unrelated articles that follow the same underlying train of thoughts.
Problem is, there are at least 2 or 4 separate such trains rushing through my mind any given week, so I'd still need some sort of "tracks" on my blog to sort these out.
I guess one possibility here is to write some articles in advance, but have a release schedule that allows me to put out closely related articles. I also need to better index old works, to some extent I've started doing that already.
One thing I will commit to, for now, is making the comment section as visible as possible, even though it pains me greatly I have to move the reactions at the bottom (and design suggestions that would allow me to keep both are welcome).
The other thing I will tentatively commit to is releasing two articles every week for the next month, one Wednesday and on Saturday. This might be a bit of an overkill and I'll probably tone it down to every Wednesday for the next few months, with some extra Saturdays thrown in, but we'll see.