Sebastian Karcher | Lightnews

Simon Donner @simondonner.bsky.social · 12h

A truly gonzo Executive Order from the Trump Administration removes US from not just the UN Framework Convention on Climate change, but also the IPCC (!), the IUCN, the IPBES (the IPCC of biodiversity), and all sorts of other organizations. www.whitehouse.gov/presidential...

Withdrawing the United States from International Organizations, Conventions, and Treaties that Are Contrary to the Interests of the United States

MEMORANDUM FOR THE HEADS OF EXECUTIVE DEPARTMENTS AND AGENCIES By the authority vested in me as President by the Constitution and the laws of the United

www.whitehouse.gov

Sebastian Karcher @adam42smith.bsky.social · 21h

ChatGPT is obv a bad for this.
Here's the thing, though: It's pretty hard to get good results for this type of q from traditional databases. Try it in Pubmed!
Otoh, using the right LLM-based tool, e.g. Elicit, produces good results with a natural language query (tho I'd worry about the summaries):

Elicit search results, showing first 5 papers to the query "Are there any case studies of implanting defibrillators shortly after heart surgery"

Sebastian Karcher @adam42smith.bsky.social · 23h

That "Why?" somehow hits different for "Let's draw some causal graphs"...

Sebastian Karcher @adam42smith.bsky.social · 1d

Characteristically excellent from Robin. Some of these (like 2&5) are things FLOSS/open infra folks have been saying for a while; might be worthwhile considering reasons for slow progress.
Others (esp 4) are new to me & likely more controversial

Robin Berjon @robin.berjon.com · 1d

I'd like to see some coordination in the open source community to respond to this. The outcome could shape significant funding and strategic moves.

Off the top of my head, there's a few things that I'd like to see in a new European open source strategy. 🧵

Maximilian Henning @maximilian-henning.bsky.social · 1d

The @ec.europa.eu will soon present “a combination of funding and policy measures” to support European open source projects in becoming commercially viable alternatives to proprietary US tech.

Read my article at @euractiv.com:

www.euractiv.com/news/commiss...

Kathleen Weldon @kathleenweldon.bsky.social · 1d

When this guy wrote his survey of European polls in 1963, he had no idea that he would be thrilling an archivist in 2026. But these charts warm the cockles of my heart.

List of standard background variables and number of survey organizations saying they were covered always or often, by region within Europe. Age/sex/occupation were basically universal, while party membership, frequency of religious attendance and vote in last election were rare. Perhaps surprisingly given current practices, household income was fairly common, but far from universal

Chart of survey of polls from European nations in 1963, grouped by region (Scandinavian, Benelux, Austria/Germany, France/Italy/Switzerland, Other) Chart provides number of total interviews as reported and also as calculated, based on number per survey, compared to population, showing that the interview per thousand inhabitants was highest in the Benelux and Scandinavian countries (58, 60) versus 13 in Austria/Germany and 9 in France/Italy/Switzerland.

Robin Berjon @robin.berjon.com · 1d

I'd like to see some coordination in the open source community to respond to this. The outcome could shape significant funding and strategic moves.

Off the top of my head, there's a few things that I'd like to see in a new European open source strategy. 🧵

Maximilian Henning @maximilian-henning.bsky.social · 1d

The @ec.europa.eu will soon present “a combination of funding and policy measures” to support European open source projects in becoming commercially viable alternatives to proprietary US tech.

Read my article at @euractiv.com:

www.euractiv.com/news/commiss...

Commission wants to commercialise open source for sovereignty | Euractiv

An incoming strategy looks set to target support at European open source projects to scale into commercial alternatives to US tech

www.euractiv.com

Richard McElreath 🐈‍⬛ @rmcelreath.bsky.social · 1d

This is traditional!

Sebastian Karcher @adam42smith.bsky.social · 1d

This is a delightful story as well as impressive science & I love @carlzimmer.com for including this bit (gift link):

www.nytimes.com/2026/01/01/s...

Screenshot of text with the last sentence (starting with "the elephant kills") highlighted

Americans by no means have a monopoly on the game. People play it around the world in many variations, and under many names. In Japan, where the game has existed for thousands of years, it’s known as janken. In Indonesia, it’s known as earwig-man-elephant: The elephant kills the man, the man kills the earwig and the earwig crawls up through the elephant’s trunk and eats its brain.

Sebastian Karcher @adam42smith.bsky.social · 1d

I do know people who take their car to the race track, so this isn't completely out of the question, but probably something that could be addressed.

Sebastian Karcher @adam42smith.bsky.social · 1d

The guys really, really love their calipers...

Carl T. Bergstrom @carlbergstrom.com · 1d

"The application...asks users for their height, ancestral background, SAT score and their feelings about entrepreneurs....The system can assess their cheekbone prominence, jaw strength or body-fat percentage from a scanned photo, and analyze their application to estimate an IQ score."

Sebastian Karcher @adam42smith.bsky.social · 1d

Same here. I was surprised that's the direction in which they went (didn't make it to the actual ideas, survey link timed out).
Is this different in other fields, perhaps? Like medical researchers don't know what to try next?

Carl T. Bergstrom @carlbergstrom.com · 1d

"The application...asks users for their height, ancestral background, SAT score and their feelings about entrepreneurs....The system can assess their cheekbone prominence, jaw strength or body-fat percentage from a scanned photo, and analyze their application to estimate an IQ score."

Sebastian Karcher @adam42smith.bsky.social · 1d

Yeah, it's very easy to add something to (potential) training data in the chat interfaces. How/how easy to opt out varies by tool (not sure it's even possible for all free ones). Enterprise versions (i.e. a tool your institution provides access to) are typically opted out by default, as are APIs.

Sebastian Karcher @adam42smith.bsky.social · 1d

It does say "custom AI models" which strongly implies that, but it's one thing they should, certainly in retrospect, have made clearer -- writing emails like this is super tricky bc they can't be too long/complicated.

Sebastian Karcher @adam42smith.bsky.social · 1d

The boring answer is stand mixer & espresso machine.
The two w the most unexpectedly large impact are pasta roller attachment for KitchenAid & digital kitchen scale

Kari Dru @karidru.bsky.social · 2d

Everybody has two kitchen appliances that will change their life. The problem is, they're different for everybody.

Michael Druker @mdruker.app · 2d

This place needs some Innocuous Discourse pronto. Quote this with a take that’s not political or aggressive

Sebastian Karcher @adam42smith.bsky.social · 1d

Since several people have complained to the IRB, I'd assume that's the angle this gets resolved under. FWIW, I'm quite convinced that there, too, the researchers are easily in the green legally.
Beyond legality, I find the ethical case against this work weak & problematic, but worth discussing.

Sebastian Karcher @adam42smith.bsky.social · 1d

I think that's right for fair use - as I say, the boundaries are blurry. But I don't think what the researchers here did even requires fair use.
Analyzing a paper with a computer model doesn't qualify as "reproduce", "prepare derivative work of," or "distribute" (the uses protected by copyright)

Sebastian Karcher @adam42smith.bsky.social · 1d

I _love_ that you're doing this

Sebastian Karcher @adam42smith.bsky.social · 1d

#datalibs smoke signal!!!

Conrad Hackett @conradhackett.bsky.social · 1d

Please share - @pewresearch.org wants to hire a data archivist who will be an advocate for data users, helping to ensure that our datasets are easy to discover and reuse by researchers, journalists, and the public.
pewtrusts.wd5.myworkdayjobs.com/CenterExtern...

Pew Research Center is seeking a Data Archivist to support our commitment to open science and data transparency. This newly created role will play a key part in enhancing the accessibility, usability, and reproducibility of our research data while continuing to protect the privacy and identity of our survey participants.

As Data Archivist, you will lead efforts to create and implement best practices for preparing, documenting, and disseminating datasets. These best practices should maximize FAIR (Findable, Accessible, Interoperable, and Reusable) principles while minimizing disclosure risk. You will work across teams to ensure our data is well-organized and thoroughly documented.

You will serve as an internal advocate for data users, helping to ensure that our datasets are not only accurate and comprehensive but also easy to discover and reuse by researchers, journalists, and the public.

This is a full-time, Pew Research Center position. The position is funded by an external grant and limited to a two-year term.

Primary Responsibilities

Develop optimization procedures to improve discoverability of our datasets on internal and external platforms

Develop and maintain standards to improve accessibility of our microdata and tab plans by changing/adding formats and/or adding documentation

Identify metadata documentation best practices and a process to implement those best practices at the Center

Work with Legal to evaluate most appropriate license to publicly share the Center's survey data, including Creative Common options

Identify and correct processing inefficiencies in our data publication process

Sit on the internal Disclosure Risk Taskforce

Document analytical decisions and code to support transparency and replicability, including the development of a RACI chart for publishing code to recreate derived variables that are used in reports but are not included in the microdata

Manage/create merged time series datasets for select Center datasets

Identify a process for internally archiving data and projects that are no longer in active use

Identify and implement a process to assign Digital Object Identifiers (DOI) to microdata

Prepare and upload public-facing datasets and restricted-use datasets for external sharing.

Train staff on FAIR principles and best practices in data archiving.

Education/Training/Experience

Bachelor’s degree required, preferably in library sciences, organizational management, or a related field.

5-7 years of experience with data archiving, database management, or survey research. This may include graduate training at the MA/PhD level or equivalent experience in an applied setting.

At least 3-5 years of experience applying FAIR and open science principles.

Background in social science research or data curation.

Experience in data management, archiving, or research support.

Familiarity with FAIR principles, Creative Common licensing, data privacy principles, and exposure risk.

Proficiency in metadata standards and documentation tools.

Experience managing research projects, including working collaboratively with a multidisciplinary team.

Experience with statistical software (e.g., R, Python, Stata) and reproducible research workflows.

Knowledge, Skill and Workplace Requirements

Strong organizational and communication skills.

Detail oriented with exacting standards to maintain accuracy and impartiality in all work products.

Ability to work independently to carry out special projects from start to finish.

Ability to balance numerous tasks simultaneously.

Ability to work collaboratively and collegially with other team members, as well as with staff from other Pew Research Center teams.

Ability to balance competing priorities and identify optimal solutions

FLSA Status: Exempt

Compensation: Starting salary is commensurate with experience within the range of $100,000 - $120,000.

Hybrid Work Schedule: Pew Research Center staff are required to be present in the Center’s Washington, D.C., office three core days weekly (Tuesday, Wednesday, Thursday). Staff may work virtually from remote locations on other days in a typical work week.

Conrad Hackett @conradhackett.bsky.social · 1d

Please share - @pewresearch.org wants to hire a data archivist who will be an advocate for data users, helping to ensure that our datasets are easy to discover and reuse by researchers, journalists, and the public.
pewtrusts.wd5.myworkdayjobs.com/CenterExtern...

Sebastian Karcher @adam42smith.bsky.social · 1d

I mean -- copyright licenses are, in fact, useless to ban things that copyright law either doesn't cover (like algorithmic analysis) or exempts (like fair use). And again, that applies equally to preprints and gated papers.

Sebastian Karcher @adam42smith.bsky.social · 1d

My disagreement is that it's not their right. As a copyright holder, you don't have the right to prohibit algorithmic analysis of your work, which is all that happened here.

They likely meant to be extra safe & exclude works that weren't explicitly licensed for re-use & screwed that part up.

Sebastian Karcher @adam42smith.bsky.social · 1d

Fair use generally applies to gated publishing, too, as long as you obtain the papers legally (that's why Anthropic readily paid to libgen authors). The case for fair use is marginally weaker, but not really.

Sebastian Karcher @adam42smith.bsky.social · 1d

I don't know what you think that shows? I understand some people didn't have permissive (e.g. CC) licenses on their preprints. It might have been better had those papers be skipped & the mention of a license in the email is confusing.
Fair use still applies and you don't need to 'claim' it.

Sebastian Karcher @adam42smith.bsky.social · 1d

That doesn't make sense. Fair use is an exemption to copyright (i.e., allows the use of copyrighted work w/o prior permission) and thus by definition cannot be prohibited by or violate a copyright license.
(And analyzing a paper algorithmically doesn't touch upon copyright at all).

Sebastian Karcher @adam42smith.bsky.social · 1d

3/3 What bothers me about this whole thing is that folks hate LLMs so much that they readily throw overboard all concerns about regulatory overreach into research via IRBs or hampering research via copyright. There are long, long literatures on both of these things!

Sebastian Karcher @adam42smith.bsky.social · 1d

2/3 If they trained an LLM on it (it doesn't sound like they did), they're on much weaker ground ethically, but legally, given it's non-profit research and uses papers shared at no cost, the Fair Use case would be strong in the US and TDM exceptions likely apply elsewhere.

Sebastian Karcher @adam42smith.bsky.social · 1d

IANAL, but depending on the details, I don't even think the license matters. If all they did was use an existing LLM to generate "research ideas" based on a legally obtained paper, that's just not affected by copyright at all. Whether you use Acrobat to read it or an LLM makes no difference 1/3

Sebastian Karcher @adam42smith.bsky.social · 1d

plain-text.co
source:
github.com/kjhealy/plai...

The Plain Person’s Guide to Plain Text Social Science

plain-text.co

Sebastian Karcher @adam42smith.bsky.social · 1d

a) not by federal standards: if I publicly disseminate my analysis criticizing someones work, that almost certainly meets standards for "research" (just not for human subjects).
b) lots of papers re-analyze data for other purposes than audit, so if you don't like the Gino example pick a diff one

Add to Home Screen

Light up
your news

Add to Home Screen

Light upyour news

Sign in to Lightnews

Sign up to start reading

Connect Bluesky

Connect with Bluesky

Light up
your news