This is becoming a bit of a problem for Nature titles...
retractionwatch.com/2026/03/06/l...
07.03.2026 16:58
👍 30
🔁 7
💬 2
📌 0
This is becoming a bit of a problem for Nature titles...
retractionwatch.com/2026/03/06/l...
Some people showcase intelligence not by discussion but by playing a game called "find the flaw". It is a very easy game but irresistible to junior paper reviewers, critical theorists, and terminal posters. If you know who's playing, it explains everything from reviewer 2 to "you hate waffles".
« Wikipédia est centrale dans le fonctionnement de la démocratie numérique à l’heure actuelle, mais peu s’y intéressent et presque personne ne sait comment ça fonctionne »
repost.hypotheses.org/19676
A Swedish state-funded research funding agency, Formas, has started to approve applications based partially on random draws. Applications are grouped into high quality (gets funding), low quality (doesn't get funding) and a middle group where a lottery takes place 🎰.
formas.se/en/start-pag...
Thank you @nulliusinverba.bsky.social for episode 76. I cannot agree more. We need funding by lottery and we need to incentivize teams, not individuals.
We should cap the number of articles submitted to arXiv at 5 papers per year per (co-)author.
Are #openaccess fees that some journals charge authors too high? The Chinese Academy of Sciences, the world’s largest research institution, reportedly thinks so and plans to stop funding some, a move that could shake up #scientificpublishing. @science.org www.science.org/content/arti...
Helmholtz Open Science Office CoARA action plan published to implement the ten commitments of the agreement on reforming research assessement
#CoARA action plan published!
🕶️ Overview of the interactions between open science and #ResearchAssessment reform
👁️ Classification of the current status quo
🚀 Measures and planned activities for the period 2026–2030
👉 doi.org/10.5281/zeno...
@coarassessment.bsky.social #OpenScience
1/
New paper: Do LLMs just predict the next token or can they plan ahead?
Previous work (Anthropic) suggested Claude Haiku plans ahead when writing poetry - but that was one model and a few prompts.
We found quantitative evidence that many models engage in implicit planning 🧵
AI makes continuous reproducibility and robustness testing trivial. What happens to science under new levels of scrutiny and stress-testing by default?
Some thoughts on how this could play out, informed by watching open science play out over the last decade.
Il s'agit en l'occurrence de *recommandations*. La seule *obligation légale* est de déposer un document qui parle de ce que l’on a fait au cours des 5 dernières années. Donc arial ou n'importe quelle autre police, c’est tout pareil.
Que dit le "décret" ?
J’ai utilisé New Computer Modern, sans serif, pour avoir la phonétique et le cyrillique
Je conseille la lecture de ce texte à toute personne qui s'intéresse à la recherche scientifique :
web.archive.org/web/20100401...
Note: tech companies are happy about these shortages, because their next play is renting hardware out to people for a subscription fee.
Puisqu'on est dans l'actualité de niche ESR, on m'a informé aujourd'hui qu'Étamines va disparaître. Ne soyez pas trop tristes…
Guy Parmelin exclut les sciences humaines des fonds de recherche nationaux
www.tdg.ch/parmelin-les...
📄 New paper:
Pre-Editorial Normalization for Automatically Transcribed Medieval Manuscripts in Old French and Latin
Thibault Clérice, @rachelbawden.bsky.social , Anthony Glaise, Ariane Pinche, @dasmiq.bsky.social (2026) arxiv.org/abs/2602.13905
We introduce Pre-Editorial Normalization (PEN).
🧵⬇️
Piratage de données RH CNRS
Des données personnelles d'agents CNRS rémunérés avant 2007 ont été "téléchargées de manière non autorisée"
Plus d'infos ici : www.cnrs.fr/fr/exfiltrat...
Par contre, on ne sait pas de quand date cette "exfiltration" ?
#IA Sécurité des données : le déploiement du chatbot du CNRS par Mistral n’est « pas sérieux » next.ink/223959/secur... par @mart1oeil.eu #Mistral
Graphs of sensitivity, showing LLMs outperforming humans
We coded our ~100k articles using LLMs. Should you believe them? To answer this, we benchmarked 4 human RAs against 3 LLMs on their ability to recover ground truth article data. Details in the paper and appendices, but the LLMs did well and handily beat the highly trained humans.
« Quand @awscloud.bsky.social ou Azur tombera, la communauté éducative peut ne plus du tout s’en apercevoir »
Savez-vous que @education.gouv.fr grâce aux communs numériques, est plus robuste aux aléas géopolitiques que la plupart des grandes entreprises françaises ?
vimeo.com/1161269381
The only real solution is to significantly increase the number of positions open for competition.
Normalized h index, k index, m index, normalize h index for age/seniority, normalize IF for field, for subfield (nuclear chem ≠ materials chem), … I’m not sure it’s a good road (and all indices you can create will be gamed)
La demande en #IA et IAg en #SHS est telle, que je me retrouve à refuser des demandes de participation à des projets de logiciels, services, ANR, etc. afin de faire attention à ne pas transformer le HN Lab en startup et moi en manager de fiches de temps. Là aussi il y a une bulle !
HAL is great, but it cannot be compared to arXiv. It is not used by non-French scientists. Furthermore, editorial practices in the humanities and social sciences are very different from those in mathematics. The submission of preprints is exceptional in linguistics.
unfortunately, arXiv does not accept preprints of linguistics papers
The other difficulty is that a jury of 18 people, renewed every four or five years, is not able to gain a sufficiently accurate understanding of the candidates' actual contributions. Unfortunately, this also applies to promotions.
In my opinion, DORA is not compatible with competitive recruitment. How can you compare contributions in subfields with completely different practices?
sigh.......
"Scientists who engage in AI-augmented research publish 3.02 times more papers, receive 4.84 times more citations and become research project leaders 1.37 years earlier than those who do not."
from
www.nature.com/articles/s41...