Joel Chan

@joelchan86

(Associately??) Professing #hci #creativity #toolsforthought #metascience at UMD INFO and HCIL #immigrant (🇲🇾 ➡️ 🇺🇸 ) #firstgen academic Building @discoursegraphs.bsky.social #atproto 🤝 #openscience

1,261 Following1,947 Followers243 Posts

Bluesky

Is this your profile?Claim it and joelchan86.dirty.social is yours.Claim profile

Joel Chan@joelchan86· 2d

One of my favorite features on the Lea app: collating discussions of papers! Beautiful companion to @semble.so for social sensemaking of science. Check it out! https://bsky.app/profile/did:plc:5wm25vgenhgut3iqfjf4ozj5/post/3msvs2hxmvc23

screenshot of discussions UI on the Lea app that collates Bluesky posts and threads that mention a research paper URL

Joel Chan@joelchan86· 11d

h/t @semble.so discovery trail via @zafarali.me

Joel Chan@joelchan86· 16d

Joel Chan@joelchan86· 17d

@semble.so collections of collections???

Joel Chan@joelchan86· 19d

Yes! Everyone should know about Public Lab and TREEKit! lizbarry.net Mea culpa for forgetting to add to the Semble Collection! (just added!)

Joel Chan@joelchan86· 24d

Joel Chan@joelchan86· Jun 24

I too welcome multiple layered indicators, but emitting a single ~1% ranking/inclusion QED score seems unlikely be helpful for this purpose. Would prefer at least the granular separated novelty / supportedness indicators (but even then I'm skeptical this can be collapsed across fields).

Joel Chan@joelchan86· Jun 19

yay my #atmosphereconf shirt finally arrived! channeling this energy into the summer!!

Joel Chan@joelchan86· Jun 19

Just pushed a bunch of cards to my @semble.so collection via the API from my bookmarks! thank you opus and thank you @cosmik.network team for the API! tab sanity 🤝 contributing to the knowledge commons 🎉

screenshot of a code editor - on the left is code for interfacing with the Semble API; on the right is a list of links in a .tsv file to be uploaded to Semble

a screenshot of items in a Semble collection

Joel Chan@joelchan86· Jun 11

methods: datawrapper viz of data from OECD: www.datawrapper.de/_/iQQ0V/ data-explorer.oecd.org

Joel Chan@joelchan86· Jun 11

Analysis of OECD data showed that the number of researchers (and published research outputs) has been growing while the number of scientific journalists who can translate this to the public is simultaneously dropping at similar rates #evd

Joel Chan@joelchan86· May 17

can't see this specific record, but i'm VERY intrigued by the rest of the synthesis islands!! cc @infotainment.bsky.social as a potential output format for our living synthesis

Joel Chan@joelchan86· Apr 30

Meanwhile on LinkedIn... (tbf there is some good stuff there, but have to do a lot to sort the stew from the slop)

linkedin post criticizing ai slop but is ai slop itself!

post text follows below:

I got tired of the AI slop flooding LinkedIn. “Most leaders don’t realize...” “It’s not about X, it’s about Y.” Same structure, same cadence, same nothing.

People use AI to write content, fine.

But it doesn’t have to sound like everyone used the same prompt.

Here’s the thing though: the problem is real.

Leaders have opinions worth hearing. They see trends before anyone else. They have takes that would actually be worth reading. But writing takes time they don’t have.

So the content doesn’t happen.

Or it gets outsourced to someone who makes it sound “professional”, which means it sounds like everyone else.

We built Edgar to fix that.

An AI calls you once a week. Interviews you for 10 minutes. Asks about your work, your opinions, what’s been on your mind. Then writes LinkedIn posts that actually sound like you.

Your tone, your vocabulary, your way of saying things.

You approve. Edit if you want. They go straight to scheduling.

We’ve been testing this for a few months. The feedback: it feels like a ghostwriter who already knows how you think.

Beta is open. If you’re a leader who should be posting but never finds the time or you run marketing and need more executive visibility, come try it.

https://edgarwrites.ai

author of post admits to writing the post with ai: "yes the post is made with early prompt version, and it does look like ai slop to be honest. Refinements have meen made tho 👍"

Joel Chan@joelchan86· Apr 21

From post-publication review to post-publication enrichment 🔥🚀 #ATScience #Metascience

Quote:

This is the deeper promise of facets on a decentralized network: publications become living objects that can be enriched over time by a distributed community, without requiring the original authors or the hosting platform to do anything. The original text stays immutable in the author’s or preprint-servers PDS. Annotations accumulate as separate, signed records from other identities. Consumers choose which annotation layers to display. It’s a model that could make post-publication enrichment — RRIDs, CiTO intents, corrections, links to reproducibility artifacts, lay summaries — a natural part of how scientific documents live on the network, rather than something bolted on and owned by a single platform.

Joel Chan@joelchan86· Apr 18

main case study in the paper is of predictions of "replicability" of studies!

Joel Chan@joelchan86· Apr 18

#TIL new term! "inference by false ascent" a complement/alternative to goodhart's law for understanding the dangers of metric-izing research evaluation from: arxiv.org/pdf/2601.09753 #metascience

Quote:

In the past, scientific communities – when faced with increased pressure to evaluate and sort
research and their authors – have decontextualized, repurposed, and aggregated newly crafted credibility
markers (Lee 2022) through fallacious processes of inference by false ascent. In inference by false
ascent (Cartwright et al. 2023, 48; Reiss 2016), the original, concrete purpose for which a measure is
originally designed gets redescribed at a more abstract level, but in epistemically problematic ways: for
example, although the journal impact factor was first introduced for the purpose of helping librarians
decide which journals to purchase (Gross and Gross 1927), some embraced this metric as a more
abstract measure of “scientific impact” (Garfield 2006). Because the more abstract purpose covers a
range of other more concrete purposes, the original measure may be deployed for those other concrete
purposes, even when doing so is epistemically problematic: for example, the journal impact factor, once
elevated into an abstract measure of “scientific impact,” was repurposed as a controversial measure of
article and author impact (Hicks et al. 2015; Larivière et al. 2016) and likely informs determinations of
article relevance in research search engines like Google Scholar (Google Scholar, n.d.). Even open science elements shared for the purpose of facilitating the transparent evaluation of a manuscript’s claims
(Begg et al. 1996; Nosek et al. 2015) can be easily decontextualized and repurposed (Leonelli 2023) to
score manuscripts for their computational reproducibility (Siegel et al. 2024) and, when combined with
data citation metrics, to quantify author data-impact via “S-index” metrics (Challenge.gov, n.d.).

Joel Chan@joelchan86· Apr 10

This is my clumsy interpretation of the lovely vision painted by @maggieappleton.com of "home-cooked software and barefoot developers": maggieappleton.com/home-cooked-...

Diagram arguing that small, niche software built at low cost represents an underexplored 'land of opportunity' compared to the dominant model of high-scale software. 2x2 matrix with Scale on the x-axis (from 'Small, Special Snowflake Software' to 'Global Industrial Software') and Profit on the y-axis (from 'Sink Hole' to 'Billion $ Unicorn'). Three zones are highlighted: 'Most software today' sits in the high-scale, high-profit quadrant. 'Short lived failures' occupies the high-scale, low-profit quadrant. 'Land of opportunity' sits in the low-scale, medium/low-profit quadrant, labeled as low cost. The low-scale, high-profit quadrant is empty."

Quote from Ivan Illich's "Tools for Conviviality":

"People need not only to obtain things, they need above all the freedom to make things among which they can live, to give shape to
them according to their own tastes, and to put them to use in caring for and about others"

Joel Chan@joelchan86· Apr 10

This (+ things like Lea: atmosphereconf.org/event/ats26-...) is what excites me about the opportunity from AI-assisted coding: making awesome software-enabled things exist separate from the pressures and incentives of VC

Photo of Maria standing next to a projected slide titled 'The origin story.' The slide describes how Maria and Eugene are both active on social media and in the AI research community, frustrated by colleagues' reluctance to leave X. It notes Eugene faced mass bullying for his AI posts. A red box highlights the key line: over winter break, they both got obsessed with Claude Code and found energy to attempt something to fix their bleak social media landscape.

Joel Chan@joelchan86· Apr 10

hell yeahhhh tons of synergies with what we're doing @discoursegraphs.bsky.social too (we're already collab-ing with @semble.so)

Quote:

"Pivot to Wiki" is my proposal for a cultural shift in how we organize, curate, and engage with information. we need to rely less on info formats that work by grabbing our limited attention. we've only got 24 hours in a day, and all the articles & social media posts & flashy videos have to compete for each second of that. when clicks and retention rule, conveying truth becomes an afterthought.

consider the financial incentives for article-based news/journalism & social media:

article-by-article journalism, as a business, must necessarily be written by a finite number of employees and derive its revenue from keeping readers informed on the news via clicks or subscriptions. if it stops producing news (new information), it won't produce any revenue.
social media, as a business, has to attract a mass audience who constantly post to keep each other's attention. if nobody posts, there's no business.
i'm not suggesting we reject news and social media entirely, but you need to realize at a very basic level that these organizations, without a change in what we demand, cannot escape the attention competition.

we as consumers have to demand that as much information as possible be converted to and represented in Wiki formats. a Wiki, compared to a news publication, sorts information not by time but by topic. a Wiki prioritizes interlinkedness and organization, and derives its value from those properties.

Joel Chan@joelchan86· Mar 27

neat way to conference in an interop way - see neat comment in @stream.place , directly go follow that user on @bsky.app #atmosphereconf

Joel Chan@joelchan86· Feb 27

LOVE this vision, and how @semble.so's upcoming connections feature helps to build towards this. Open, accretive collective sensemaking FTW!! Seize the means of sensemaking!! leaflet.pub/p/did:plc:y3...

Screenshot of text:

From feed to document: This is the direction that feels more novel: the idea is that posts can be associated with an existing piece of writing. For example, if I posted this Leaflet and a month later, a post discussed a related need, a similar project, or a desired feature, somebody could "plant" (or associate) that post into this document.

That gesture feels different to me than adding the post to Are.na or Semble. A collection is flat, and the blocks are collected because they fit under the same umbrella. On the other hand, a piece of writing already has structure, whether it's making an argument, enumerating a list, or telling a story. Planting a post into a piece of writing allows the editors of the document to later easily synthesize the new information.

Joel Chan@joelchan86· Feb 12

Lovely vision of decentralized networks of validators, from @trfetzer.com: margin.at/annotation/a...

Screenshot of text:

Is that we move to community based AI with essentially a decentralised network of validators of information creating trust web and nodes that confer community trust. You can validate, you can verify. And ultimately, you may also be authenticated to nudge and “manipulate” but in an environment of care and around loving and positive interpersonal relationships. This may be naive but we each carry a device with us that could become part of this network and we each carry a network of friends connected in the digital realm that may contribute to information validation services.

Joel Chan@joelchan86· Feb 7

More people should know about the amazing story of Tim Friede. And the world should have far more of these stories. THIS IS SCIENCE. I am curious, though, why Tim isn't credited as a direct contributor to this resulting pub (though he is at least acknowledged): www.cell.com/cell/abstrac...

Screenshot of text:

Tim Friede is an autodidact herpetologist and venom expert. Almost twenty years ago, while collecting a home collection of venomous snakes, Tim began self-administering diluted venom as a means of establishing immunity in case he was ever bit accidentally. Over the course of nearly 20 years, he self-administered over 700 escalating doses of snake venom from the world’s deadliest snakes, culminating in the ability to be bitten by cobras, taipans, black mambas, rattlers and other venomous snakes and survive (he has now been bitten over 200 times). Realizing that he had achieved a level of hyperimmunity that was unusual for a human, he began reaching out to the therapeutic community asking to be researched in order to generate a universal antivenom. In 2017 he and Jacob Glanville were put in contact, and begun the collaboration that has ultimately given rise to the successful discovery of dozens of broad-spectrum anti-venom antibodies from his blood and the awarding of a National Institute of Health SBIR award to develop the polyclonal as the world’s first fully human broad-spectrum antivenom.

Screenshot of publication title ("Snake venom protection by a cocktail of varespladib and broadly neutralizing human antibodies") with contributor list that does not include Tim Friede

Screenshot of acknowledgements section of a paper that shows credit being given to Timothy Friede for blood sample donations and detailed knowledge of herpetology.

Joel Chan@joelchan86· Feb 5

Common experience reading text on the Internets of 2026: > See interesting headline > Open the thing > Read first few lines > "It's not X... it's Y. It's not X, it's Y." > Close the thing

Joel Chan@joelchan86· Feb 1

No more error detection even now?

Joel Chan@joelchan86· Jan 31

i digged @moja.blue's enumeration of some familiar anchors here (galleries evoke curation and commons-ish, which i like) plurality.leaflet.pub/3mdkt73irvs2m

Joel Chan@joelchan86· Jan 16

Some interfaces that play well with this too: Dreamsheets (exploring rotations in latent image space: doi.org/10.1145/3613...), and Patchview (exploring rotations in latent "narrative world" space: doi.org/10.1145/3654...)

Joel Chan@joelchan86· Jan 11

Really fascinating to watch progress on AI-assisted research in mathematics, especially since we have an incredible narrator in Field Medalist Terence Tao! #TIL he's been curating examples of attempts/successes here: github.com/teorth/erdos...

screenshot of a wiki about AI contribution to erdos problems. it shows the description of the wiki (collecting "various ways in which Ai tools have contributed to the understanding of Erdos problems"), as well as various disclaimers about how to interpret the contributions listed.

Joel Chan@joelchan86· Jan 8

Love it. Seeds of continuous science!!

Quote screenshot:

I hope that a lasting contribution of my work is the open, modular approach that I have curated and the community that I have helped to seed around these ideas.

Joel Chan@joelchan86· Jan 7

Thinking about analogy of "recycling waste information" re: distributed new knowledge infrastructures for science, from Paul Edwards' classic paper: journals.sagepub.com/doi/10.1177/... (h/t @infotainment.bsky.social) #commonknowledgefutures