Science Is Still Persuasion
이 글은 AI가 과학 연구 속도를 획기적으로 높이면서 연구자의 선택에 따른 불확실성, 즉 '연구자 자유도'가 과학적 설득력에 미치는 영향을 논의합니다. 표준 오차 외에 연구자가 데이터 처리, 분석 방법 등에서 내리는 다양한 결정들이 결과에 미치는 영향을 비표준 오차로 개념화하며, AI가 이를 빠르게 탐색하고 보고하는 새로운 과학적 검증 방식을 가능하게 한다고 봅니다. 저자는 향후 논문과 함께 분석 도구를 제공해 독자가 직접 '만약에' 시나리오를 실험할 수 있는 AI 기반 플랫폼이 등장할 것으로 전망합니다. 이는 과학적 재현성과 신뢰성을 높이는 한편, 학술 출판의 역할 변화도 예고합니다.
https://hunchbox.substack.com/p/science-is-still-persuasion
#ai #scientificmethod #researcherfreedom #uncertainty #reproducibility
🔐 Debian 14 verrouille sa sécurité : vous allez pouvoir compter sur "Forky" !
https://goodtech.info/debian-14-forky-reproducible-builds-security/
> Fini les "boîtes noires" logicielles. Debian bloque désormais tout paquet binaire qui ne peut pas être reconstruit à l'identique. L'intégrité avant tout.
🍥 Debian Now Blocks Non-Reproducible Packages from Testing
｢ The main benefits include package integrity, build transparency, and stronger supply-chain verification. This way, Debian can more effectively confirm that distributed binaries match their source, detect unexpected build changes, reduce the risk of compromised infrastructure producing altered packages, and prevent reproducibility regressions from entering testing ｣
https://linuxiac.com/debian-now-blocks-non-reproducible-packages-from-testing/
This coming Wednesday (!) 13 May ReproducibiliTea in the HumaniTeas is delighted to welcome Andrey Anderson dos Santos to speak about Open Peer Review. After a short input talk, we will have plenty of time to discuss this rather controversial topic!
Join us at 16:00 CEST in room 4.006 of the @unibibkoeln for tea and cookies, or online via Zoom (see link sent via our mailing list: https://lists.uni-koeln.de/mailman/listinfo/reproducibilitea-humaniteas).
Andrey recently completed his PhD on #OpenScience practices in the #humanities and is coming all the way from Brazil to Cologne, so don't miss out!
Recommended (optional) reading for preparation: https://doi.org/10.1007/s11192-020-03488-4.
"As of May 9, 2026, Debian’s migration software now actively blocks packages from migrating into the testing archive if they fail the reproducibility check. This applies both to new packages that cannot be reproduced and to existing packages whose reproducibility has regressed. Debian 14.0 will be the first major Debian release to ship under this hard mandate."
https://pbxscience.com/debian-mandates-reproducible-packages-for-debian-14-forky/
Debian Mandates Reproducible Packages for Debian 14 "Forky". In a landmark policy shift, the Debian release team has made reproducible builds a hard requirement — meaning no package may enter the archive unless it can be built bit-for-bit identically by independent parties.
Debian must ship reproducible packages
https://lists.debian.org/debian-devel-announce/2026/05/msg00001.html
#HackerNews #Debian #reproducibility #open-source #software #packages #development
My interest in Guix is mainly in computational reproducibility as in reproducing published research from a few years ago, not just installing a conda environment.
That is, you need the ability to re-deploy an identical computing environment at *anytime*, whether it is by an independent research team aiming to reproduce your results the same year as you published it, or a reviewer trying to verify your computational results.
You wouldn't believe the mess that are computer dependencies, but it's probably best described by a nightmarish DAG from [1].
Happy to say that installing julia 1.8.5 is as simple as running `guix install [email protected]`!
2/2
Bob Calin-Jageman referencing extremely useful resources on experimental design and #reproducibility in #neuroscience :
Beyond the Basics: Experimental Design in Neuroscience
https://blog.eneuro.org/2026/02/discussion_beyond-the-basics-exp-design?_hsmi=417745689
sayzard
Pascal Delrot
Ethan Black
jbz
Elen Le Foll 🇫🇷 🇬🇧 🇩🇪
graste
N-gated Hacker News
Hacker News
Pascal Quach
Björn Brembs