Defender

about to send an update for "Summer of Pub" sharing initial participants in our lil @leaflet.pub blog club!

meant to do last week; finally putting together…anyone else participating? if you're joining but haven't shared your blog yet, lmk & I'll update the list!

the ~only rule = publish weekly :)

How Hank Green is contributing to the Human Memome Project A few days ago I wrote about Human Memome Project. Today I’d like to show you a concrete example of what it looks like to contribute successfully to it.

full essay here: defenderofthebasic.substack.com/p/how-hank-g...

great question!! I recently wrote an analysis/introduction to hank green from a memetics perspective (his egregore is surprisingly not well known at all in the tpot/rationalist spaces)

How memetics can go mainstream Adam Aleksic “Etymology Nerd” recently featured my work, in a video titled, “Are gay people more like spaghetti, or Antarctica?”

new essay: "how memetics can go mainstream"

defenderofthebasic.substack.com/p/how-memeti...

like, if essays started spreading that had no attribution, and no one knew where they came from, that sounds extremely dangerous when you think about it...but we already have this situation (like trying to raise people's awareness of the value of something like memetic tracing)

(1) i've been writing these essays and publishing them as screenshots (2) people keep asking "what's the source?" and I keep saying, there's no source, this IS the essay!! (3) i decided to try writing this as a way to reflect on how memes are like mini essays that have no source but no one asks

that's the way!!! like a meme!!

(stats for the "There is no source" art piece on twitter: )

"There is No Source"

would love to support this even just as a "cheerleader/rubber duck debugging" thing, keep me updated, let me know when there's a repo, let me know if I can help 😄

hell yes!!!! i've been dying to see this (and doing it using mapping libraries is the obvious choice, I want zoomable books already, like google maps but for the book). it's SO doable

it would be cool if there is a basic DB so that if I generate embeddings for like a viral Hank Green thread, with my own OpenAI key, other people wanting to look at the same thread can re-use those embeddings

are you thinking of doing a version for bluesky? I wanted to do it like Treeverse, just put in ANY bsky thread, it gets all replies, clusters them (can make users put in their own API key, or use a local model in the browser so it doesn't cost you anything)

treeverse.app

update: k-means on projected data maybe isn't that great? (I haven't tried any of that yet)

i was using PCA but someone redid my visualize with (1) UMAP and (2) k-means clustering on the projected points, NOT on the original high dimensional points, and seems to have gotten way better results

My PCA + geojson code is in this file: github.com/OmarShehata/...

epistemic shortcuts are necessary, but some increase your ability to handle complexity, and some increase the veil of complexity that allows power, abuse, corruption to fester

the input data to the map is a geojson (literally just a json with lat/lng coords). Nice thing about mapping libraries is they're made to scale (so in theory you could have like, all of blue sky like this, if you offline process & generate different zoom levels)

maplibre! (1) get embeddings (2) project to 2D (3) scale that to a latitude longitude rectangle (4) hide the underlying world map

will ping you for feedback when that's ready 😄

haha thanks!! I think this kind of tool will make a difference only if people understand it/use it/if they find real use cases for it. I'm hoping to make a version like Treeverse (where you paste in a bsky link and it generates the visual)

treeverse.app

i think another big thing is very simply the kind of systems we've already had on old internet forums: when I type a reply, I want it show me, did 10 people already say that? If so, did the OP already respond to one of them? Makes things less annoying for popular people, AND better for average user

But I am going to try preprocessing it a bit. The good news is it doesn't need to be perfect, just need to be "good enough" to show me the shape of the discourse

yeah it doesn't cut very cleanly in terms of "islands of ideas" but I think with some refining it can (like the model right now puts text that is like longer together, or stuff that is "angry" or matches emotional sentiment together, which isn't quite what I want)

like you know how, if you're a famous / notable person, you get thousands of replies, and you & I might have some valuable or insightful feedback to give, but it gets lost in a sea of people saying the same things? In this map if you have something novel to say it will be very visible

imagine you have a post with 1000 replies, but they all say some variation of (1) this is a great policy! I support it OR (2) this is terrible, I hate it for <reason X> (3) this is terrible, I hate it for <reason Y>

this map would show those 3 islands

given a viral tweet thread, we can see "the shape" of all replies. Replies that similar together conceptually appear close together on the map (planning to apply this to bluesky next)

Prep-planned scripts for social interactions are training wheels, and they prevent you from actually learning the underlying rules. You have to prompt real humans over & over to learn this. Good news: a lot of them crave earnest interaction on the internet. Go practice there.

"how autistic supercharmers develop (1/2)"

the ideological turing test is how you find out if you think the other side is evil because you don't understand them, or because they actually are evil

think of the pursuit of science like a market. Scientific paradigms compete the way companies compete. You might think those guys over there are crazy and not even doing "real science", that's fine. Until they make predictions that you fail to make, you don't have to take it seriously