Henrik Dobewall

SocArXiv releases AI policy Introducing SocArXiv’s AI policy.

We have posted our AI policy

An introduction: socopen.org/2026/03/09/s...

And the policy:
socopen.org/ai-policy/

Eine Ode auf die deutsche Sprache. In english something like socioeconomic intergenerational inequalities

The paradox of inequality that isn’t: rising economic inequality depresses and polarizes citizens’ belief in meritocracy Abstract. This study examines how rising income inequality has been impacting individuals’ belief in merit-based success, using three decades of survey dat

🔍 How does economic inequality impact beliefs in meritocracy?

Using comprehensive survey data from 39 advanced capitalist democracies over more than three decades, Markus Gangl & I examine how rising economic inequality has been shaping citizens' belief in meritocracy.
🔗 doi.org/10.1093/ser/mwag016

So it seems there has never been a d.e.i keyword list of banned terms...

Finally, this verify you are human box is making sense :-)

Many people had the early idea to make a digital copy of oneself based on published work, failed applications,code, etc. To fine-tune llm's but as we are just a drop of water in the data used to train them i would not worry too much about this digital gadget. Legally and ethically different story.

Api cost per token, self hosting is possible but you need to cover server costs. My take is that 6 month old models are still good enough for most tasks, so with a little spirit (similar to creating your own webpage) more independence at reasonable cost and performance should work out already.

Mistral AI does not currently offer a built-in “agent mode” like some other platforms, but you can build agentic workflows using Mistral’s API in combination with tools like LangChain or custom code. Many developers use Mistral models as the “brain” for their agents

Mistral AI’s models, including the one powering Le Chat, are highly competitive on standard benchmarks like MMLU, MT-Bench, and HumanEval, often outperforming many open-source and some closed-source models in efficiency, reasoning, and multilingual capabilities.

This is a very good overview on recent developments of AI and its consequences for social sciences. Outdated, of course, as of yesterday. Social scientists were always terribly bad at predicting the future, but good at shaping it through policy. And, we are used to find and define our niche.

A real problem, especially for data providers, is "system/model colapse": AI is trained on the systematic errors of AI, resulting in a situation were we are stuck in 2024/5 when knowledge was still produced by human trial and error. If i am stuck in some century then please in one with better music.

I agree, who enjoys reading 100% AI generated text. Why bother reading it. Writing academic papers is about sharing knowledge and what we have learned on the way; accelerating human capital generation. So let's keep calm and publish.

We simply don't know, for better or worse. The Western world has not seen a large GDP increase since generations. Of course, jobs will change, some may disappear, but if AI indeed results in exponential economic growth over years then this needs to be distributed (policy) and generates new jobs.

@torkildl.bsky.social there it is, the "frozen" ai agent that could potentially been used on secure servers (eg. Registers).

My take on this is that we need synthetic register data for using llm's on (not in) registers (also for training of students, open science/review, international collaboration). But until recently Statistic Finland did not really want to help or know how.Did you discuss synthetic registers in Norway?

We just had an presentation on ai in registers in FI. In NL they have already a frozen / open weight llm's on the server. "Frozen" ai agents follow soon?

Slides by Salganik: Life trajectories and life chances: New approaches from population registries and AI redir.lyyti.com/lnk/BAAACEou...

www.aeaweb.org/articles?id=...

#Econsky

Senior researcher, 4-6 positions The Faculty of Social Sciences, INVEST Flagship Research Centre at at the University of Turku invites applications for contract-based, 2-year fixed term senior researcher (4-6 positions). Employment s...

🔥 Apply for our 2-year fixed-term position as a Senior Researcher!

Open positions (4–6) at @utu.fi/INVEST in the
#STRATEQ & #USFI projects led by @janierola.net
#CoROL project led by @anttikouvo.bsky.social
#PublicPrivate project led by @mjantti.bsky.social

ats.talentadore.com/apply/erikoi...

Highly recommend research environment, and an overdue topic. New technology usually adapted faster by those better off already.

Michael, you have trained some of the most inspiring young researchers I know. The biggest risk of AI (agents or not), I see, is that it fundermentally changes how we learn new skills and communicate knowledge. We hopefully dont end up in a high productivity and low human capital formation situation

But at least in academia, most of them were delegated to researchers years back. (These skilled colleagues should not have been cut at the first place, indeed). I absolutely agree that it is outrageous that nobody seems to care about next Gen experts and how skill formation is eroding due to AI.

Sorry,for being unclear with bullshit task i meant tasks not central your job without learning curve. Some of these tasks were indeed traditionally done by specialists like assistants, e.g., booking a flight or set up a meeting.

Again, they makes the claim that it looks like that entrance level coding and project management skills are not needed any longer, while kind of admitting that all they can say is valid for 1-2 years max. Good news, "bullshit task" at work and in private life will be done by AI.

how to best interact with an AIgent (e.g., ask it what it needs to know from you to complete the task), that most code is written by code already with all it societal and policy implications.

I would recommend most people who work in knowledge jobs listening to this. It covers how AI agents taking different roles are !currently! been used (eg it was clear since years that it would take more than one AI agent to write an academic paper),

How Quickly Will A.I. Agents Rip Through the Economy? A.I. agents are here. Have they changed your life yet? The release of agents like Claude Code marked a new pivot point in the history of A.I. We are leaving the chatbot era and entering the agentic…

The conversation between Ezra Klein and Jack Clark of Anthropic is really good

pca.st/episode/130b...

THE 2028 GLOBAL INTELLIGENCE CRISIS A Thought Exercise in Financial History, from the Future

Well, in Finland we have a 10% unemployment already todAI... www.citriniresearch.com/p/2028gic

Example for the two staged unsupervised machine learning algorithm using point data as input. Backlayer maps depict Hamburg. The map shows neighborhoods in different sizes and forms, sometimes following administrative borders (black lines) sometimes not. Three differently colored neighborhood types are displayed, each representing a different social group of residents.

xample for the two staged unsupervised machine learning algorithm using 500x500m grid cells as input. Backlayer maps depict Hamburg. The map shows large neighborhoods in different sizes and forms, sometimes following administrative borders (black lines) sometimes not. Three differently colored neighborhood types are displayed, each representing a different social group of residents.

Looking for a measure of #neighborhoods, micro or macro #segregation?

I've got something for you!

My newly published paper in Sociological Methods & Research presents a machine-learning-based algorithm to delineate neighborhoods with grid-cell or point data:
journals.sagepub.com/doi/10.1177/...

We extended the deadline of our machine learning competition by a few weeks until mid March! Still time to participate. We currently have 233 participants with 111 total submissions (some of them will be test submissions etc).

Rare, dangerous side effects of some COVID-19 vaccines explained “Groundbreaking” study uncovers why adenovirus-based shots caused life-threatening blood clots and bleeding in some people

This covid vaccine research is quite fascinating (the original study is complicated to read) but in science news they explain how researchers identified thrombosis side effects and implications www.science.org/content/arti...