We're pleased to announce the Bennett School of Public Policy opening this autumn.
It will be a flagship, multi-disciplinary policy school, that will integrate & expand the teaching, research & engagement activities of the Bennett Institute.
www.bennettinstitute.cam.ac.uk/blog/bennett...
Want to learn about computational social science *for free* and launch interdisciplinary research projects? We are so excited to announce there will be *26* Summer Institutes in Computational Social Science this year! Apply to one of them here: sicss.io/locations
PhD students in social sciences studying "metascience" and AI: This looks like an *AMAZING* opportunity. Fully funded research postdoc that can be housed at whatever university you can find an appropriate mentor. sloan.org/programs/dig...
We are now looking to hire two new permanent lectureships in public policy www.jobs.cam.ac.uk/job/50248/ @bennettinstitute.bsky.social
Are generative AI tools threatening freelance jobs? Not quite, says a new study. Instead of mass job losses, demand is shifting:
๐ Fewer jobs in repetitive writing & translation
๐ Strong growth in machine learning & AI-related roles
Experience matters too.
Full story: shorturl.at/VMzZm
23/250 is Large language models reduce public knowledge sharing on online Q&A platforms
This makes me wonder if people are answering Stack Overflow questions with ChatGPT answers . . .
Thank you also to @sfiscience.bsky.social and @inet-complexity.bsky.social for organising the Complex System Approaches to 21st Century Challenges: Inequality, Climate Change, and New Technologies conference and special issue at JEBO.
๐ Read more ๐ lnkd.in/eNKRpvbj
๐ Thanks so an awesome team of collaborators!
Ole Teutloff Johanna Einsiedler Fabian Braesemann Otto Kรคssi Pamela Mishkin @oiioxford.bsky.social @csh.ac.at @bennettinstitute.bsky.social
๐๐ผ๐บ๐ฝ๐น๐ฒ๐บ๐ฒ๐ป๐๐ฎ๐ฟ๐ effects are mixed: ๐ ๐ demand ๐ด๐ฟ๐ฒ๐ ๐ฏ๐ ๐ฎ๐ฐ% and ๐๐ ๐ฐ๐ต๐ฎ๐๐ฏ๐ผ๐ development nearly ๐๐ฟ๐ถ๐ฝ๐น๐ฒ๐ฑ, but in general ๐ป๐ผ๐๐ถ๐ฐ๐ฒ freelancers in ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐บ๐ฒ๐ป๐๐ฎ๐ฟ๐ roles saw a ๐ฑ๐ฒ๐ฐ๐น๐ถ๐ป๐ฒ.
Our Difference-in-Differences (DiD) analysis reveals clear shifts in demand.
๐ ๐ฆ๐๐ฏ๐๐๐ถ๐๐๐๐ฎ๐ฏ๐น๐ฒ tasks saw ๐๐ต๐ฎ๐ฟ๐ฝ ๐ฑ๐ฒ๐ฐ๐น๐ถ๐ป๐ฒ๐โ50% for "About Us" page writing, 30% for Western language translation, and around 20% for other languagesโ๐บ๐ฎ๐ถ๐ป๐น๐ ๐ถ๐ป ๐๐ต๐ผ๐ฟ๐-๐๐ฒ๐ฟ๐บ ๐ด๐ถ๐ด๐ and ๐๐ฒ๐๐ฒ๐ฟ๐ฎ๐ป freelancers.
๐ข ๐ก๐ฒ๐ ๐ฝ๐ฎ๐ฝ๐ฒ๐ฟ alert! We document how ๐๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐๐ฒ ๐๐ ๐ถ๐ ๐ฟ๐ฒ๐๐ต๐ฎ๐ฝ๐ถ๐ป๐ด ๐ฑ๐ฒ๐บ๐ฎ๐ป๐ฑ ๐ณ๐ผ๐ฟ ๐ณ๐ฟ๐ฒ๐ฒ๐น๐ฎ๐ป๐ฐ๐ฒ๐ฟ ๐๐ธ๐ถ๐น๐น๐. ๐ผ ๐ป
๐ We analysed ๐ฏ๐ + ๐ท๐ผ๐ฏ ๐ฝ๐ผ๐๐๐, used embeddings to cluster them into ๐ญ๐ฌ๐ฌ+ ๐ณ๐ถ๐ป๐ฒ-๐ด๐ฟ๐ฎ๐ถ๐ป๐ฒ๐ฑ ๐๐ธ๐ถ๐น๐น๐, and LLMs to classify them as s๐๐ฏ๐๐๐ถ๐๐๐๐ฎ๐ฏ๐น๐ฒ, ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐บ๐ฒ๐ป๐๐ฎ๐ฟ๐, or ๐๐ป๐ฎ๐ณ๐ณ๐ฒ๐ฐ๐๐ฒ๐ฑ.
Thank you for the suggestion. We have 'unknown' categories in more recent Seshat version, so doable.
Thank you very much for the suggestion. Very creative! Should be doable with NER, though caveat might be token limits/budget. I'll notify the team. Dan Hoyer is also interested in several AI/ML applications.
Iโm looking for a new PhD student with a curious and interdisciplinary mindset to work on integrating multimodal data to better understand trajectories of mental and brain health disorders!
Deadline of applications: Feb 16!
Please spread the word.
www.jobs.ac.uk/job/DLO680/f...
Indeed, we sacrificed question readability to be able to automation 36k+ questions (though there is few-shot ex). This is also not a one-to-one comparison with a historian. I've clarified here
www.linkedin.com/posts/r-mari...
We are improving questions & prompts, please reach out with suggestions!
Thank you. Indeed, there is much to improve in the benchmark. As noted in this post, www.linkedin.com/posts/r-mari...
this is but a first step. Please reach out if you have suggestion on how to improve.
The 'sometimes' was indeed unnecessary. This benchmark was is not a comparison of with historians, it is a test of some history knowledge that requires some interpretation.
I've clarified so in this post www.linkedin.com/posts/r-mari...
We value feedback on how to improve, please reach out!
These are fair criticisms but miss the crucial point...
Academics publish in expensive OA journals because top journals and citation count remain the dominant factors in determining whether you get hired and whether you get promoted...
This will not change unless the incentive structure changes
There's an interesting new study by @ernestaigner.bsky.social @drodrik.bsky.social @jacobgreenspon.bsky.social on regional publication trends in economics - share of "authors based in low income countries remains low" disproportionate to those countries population shares vs. world โคต๏ธ
Big thanks to Fundaciรณn Sicomoro for supporting the Winter Workshop in Complex Systems with grants for students and funding our much-needed coffee breaks! โโจ Your generosity keeps the ideas (and the caffeine) flowing!
"Suspense and surprise in technological innovation"
Neat analysis of the link between expectations and outcomes in technological evolution.
Hello @bsky.app ! We are looking forward to connect with this community ๐ฆ
Make sure you follow us to see our latest news!
We'll be sharing more about our research, students' achievements, events, and everything related to #networkdatascience and #socialdatascience
#weareCEU #DNDSatCEU #seeUatCEU
Our @bennettinstitute.bsky.social annual conference is on 28th March and tickets can now be booked - in person or online: www.bennettinstitute.cam.ac.uk/events/annua... It's always a great event & we have a fab line-up of speakers/sessions
Skepticism is warranted. For robustness, we did few-shots prompting were we gave examples oh Q&A. We also repeated the test with just two categories absent/present. Performance was still poor.
This is a distinction introduced by Seshat team between those facts for which there is direct evidence and those that there isn't direct evidence but it can be inferred.
@pieterfrancois.bsky.social or Peter Turchin may be able to explain better
We are working on a follow up study.
Interdisciplinary research has many challenges and we'd love to get your feedback. Please do reach out if you have suggestions or would like to collaborate.
Thank you for pointing this out. We chose the question format for automation of 30k+ questions and to avoid biases on phrasing. It is indeed dry. Would love to get suggestions on better formatting.
To answer the second question, not yet. We'd love to for follow-up studies. Please reach out!
I agree. The 'sometimes' was unnecessary.
While the findings of this paper may seem obvious to the history community, they aren't obvious to everyone (the article is published in a Computer Science venue).
We definitely need more historians working on this :).
