... designed to collect posts and informations I found and want to keep available but not relevant to the other topics I am curating on Scoop.it (on behalf of ASSIM):
đšNew paper, now out in The Royal Society's Open Science.
How do educational videos affect people's behaviour on social media? Probably not much, but there's a massive risk of finding signal where there's only noise. Depending on how you run your analyses, you can get completely different results.
We ran two large, pre-registered field experiments on Twitter/X, to see if a previously validated "inoculation" video (about emotional manipulation), when deployed on social media, would prompt people to share less negative-emotional content.
To do so, we collected two lists of about 100,000 Twitter/X users and used the ad space to target them with either the "inoculation" video or a control video, respectively. We then scraped these users' timelines during a one-month window around the intervention's deployment, and ran emotion classifiers on their tweets and retweets. We hypothesised that the sharing of negative-emotional content and low-quality news would go down for treatment group users (compared to the control group), but not positive-emotional content or high-quality news.
Unfortunately, we didn't find any support for our hypotheses. As you can see in the first figure, there are no meaningful differences between the treatment and control groups.
Even worse is that we can't really claim a "true" null: our experiment was hampered by several design problems, including Twitter/X's "fuzzy matching" policy. This is a feature of the Twitter/X ad space (at the time anyway), where the company doesn't target the users you want to target, but rather 30% of them, with the remaining 70% being similar users Twitter/X has "matched" to your user sample according to some characteristics. This means that only 30% of our users were shown our interventions, but we don't know which 30%. In the end, we ran our analyses with about 92% noise (possibly more), making it impossible to find meaningful effects. We therefore call for social media companies to make it easier to do this type of research.
We also ran a series of additional (non-preregistered) longitudinal analyses, using different statistical frameworks. Our findings are cautionary: depending on your statistical choices and the analytical time window (looking at effects over 1h, 2hrs, 6hrs etc.), you get entirely different (and sometimes significant!) effects. In other words, without a strict a-priori analysis protocol, finding what you want to find becomes far too easy.
We therefore call on researchers to 1) create robust pre-registrations and follow them closely, being up-front about any deviations, and 2) conduct extensive robustness checks for intervention effects, before declaring positive findings.
This study was funded by Jigsaw, and written with the brilliant Jana Lasser, Malia Marks, Tianzhu Qin, David Garcia, Ramit Debnath, Beth Goldberg, Sander van der Linden, and Stephan Lewandowsky.
Information about the roles of each author of a paper can help to build trust, integrity and responsible research assessment. Coordinated efforts are needed to consolidate progress.
Ă travers des images immersives et des souvenirs qui transpirent dans la pierre et la forĂȘt, ce voyage vous invite Ă ressentir plus quâĂ comprendre.
There have been plenty of big studies in hundreds of thousands of children across the world that have thoroughly investigated the claim and found no link between vaccines and autism.
ALL evidence has been considered by public health authorities in Europe and beyond.
Only exception is for fraudolent studies, which I guess we ALL agree should be ignored, right?
Instead, what we should be mainly concerned today is the resurgence of measles.
Letâs keep our children vaccinated and protected!
If you are interested, more details from European Medicines Agency on this and other vaccines topics just published here:
Informed decisions to reduce deforestation, protect biodiversity, and curb carbon emissions require not just knowing where forests are, but understanding their composition. Identifying natural forests, which serve as critical biodiversity hotspots and major carbon sinks, is particularly valuable. We developed a novel global natural forest map for 2020 at 10 m resolution. This map can support initiatives like the European Unionâs Deforestation Regulation (EUDR) and other forest monitoring or conservation efforts that require a comprehensive baseline for monitoring deforestation and degradation. The globally consistent map represents the probability of natural forest presence, enabling nuanced analysis and regional adaptation for decision-making. Evaluation using a global independent validation dataset demonstrated an overall accuracy of about 92%.
There is no bridge between LLMs and authoritative content. That is a huge problem, and it is exactly what we are fixing.
LLMs are transforming how we discover information, but they still cannot meaningfully engage with the most trustworthy content on the internet: peer-reviewed research.
Why? Paywalls and licensing.
Right now, there is no reliable, scalable way to connect AI retrieval to authoritative research. If LLMs are the future of how we find answers, that gap is a serious problem for research institutions, businesses, and education.
Building real-time feeds of new publications across many publishers is a massive technical lift. Negotiating and maintaining licensing so AI can use that content is an equally big and still evolving challenge.
The good news is that we do not have to start from scratch.
Scite already operates indexing infrastructure that connects to leading publishers and powers our smart citations. Those citations carry context and classification and are created under direct indexing and licensing agreements with publishers.
We're working with publishers to provide to create citations for LLMs.
đ One week after Google launched Google Scholar Labs, everyone (rightfully) focused on the shiny new AI features.
But something else changed quietly â While reviewing the classic Google Scholar interface, Search Smart noticed that almost everything stayed the same⊠except one key feature: âŹïž The maximum query length increased 8-fold â from 256 characters to 2048.
This means we can now run much more complex and expressive searches that were previously impossible because of strict query caps. A major boost for power users and advanced search strategies.
â ïž IMPORTANT: Longer queries â reliable Boolean logic: Despite the new 2048-character limit, Google Scholarâs Boolean operators remain highly unreliable (OR, NOT break immediately; AND fails under complex conditions). Donât expect deterministic Boolean logic working similar to Scopus, PubMed, or Web of Science.
đHere is the full information on what you can and cannot do with Google Scholar (via Search Smart: https://lnkd.in/dcdcmKVT).
Anne-Wil Harzing đȘ â this might be especially relevant for Publish or Perish, since PoP is still the most effective tool for querying and exporting Google Scholar data.
đ§Given the timing, I canât help but wonder if the longer query limit is connected to how Labs handles natural-language questions â maybe it simply needs longer queries to work. | 11 comments on LinkedIn
Academic publishing is a very lucrative business for a very small number of private academic publishers. This paper provides a comprehensive overview of the direct and indirect channels through which public spending benefits big academic publishing companies (subscription costs, publication costs, peer reviews, scientific output). For Austria, the paper estimates that public spending (directly and indirectly) benefits publishing companies with an amount corresponding to about 25% of the annual basic funding Austrian universities receive from the Ministry of education.
To get content containing either thought or leadership enter:
To get content containing both thought and leadership enter:
To get content containing the expression thought leadership enter:
You can enter several keywords and you can refine them whenever you want. Our suggestion engine uses more signals but entering a few keywords here will rapidly give you great content to curate.
Your new post is loading...
