Sign in

Jeremy Lefort-Besnard

@jlefortbesnard.bsky.social
341 followers 638 following 91 posts

Edge AI jlefortbesnard.fr

PostsRepliesMedia
Reposted by Jeremy Lefort-Besnard
Gaël Varoquaux @gaelvaroquaux.bsky.social · 22/09/2026
Super excited about our Survey with Lihu Chen on the Role of Small Models in the LLM Era This topic is crucial in today's world direct.mit.edu/coli/article...
direct.mit.edu
What is the Role of Small Models in the LLM Era: A Survey
Abstract. Large Language Models (LLMs) have demonstrated remarkable capabilities in various reasoning tasks, which leads to the development of increasingly large models. However, scaling up model size...
0167
Reposted by Jeremy Lefort-Besnard
Cédric Herpson @herpsonc.cpesr.fr · 21/09/2026
#ESR, vers une guerre de territoire(s) ? En 2050, du fait de la baisse de la natalité, les effectifs étudiants inscrits dans l’enseignement supérieur Français auront baissé de 22%. ➡️Cela représente prêt de 700 000 étudiants de moins, dont 278 000 pour les seules universités. 1/n
499
Reposted by Jeremy Lefort-Besnard
G Milgram @gmilgram.bsky.social · 15/09/2026
🔥Nouvelle vidéo ! Ce que vous allez découvrir dépasse l'entendement. Il s'agit apparemment d'une « nouvelle façon de faire de la Science ». Un délire financé par de l'argent public et enseigné à des doctorants, en partenariat avec le CNRS😶. Juste dingue. youtu.be/oV1VNHXNizo Merci pour vos RT😘😘😘
youtu.be
Vous. Financez. Ce. Délire. 🤯
YouTube video by G Milgram
28237108
Reposted by Jeremy Lefort-Besnard
PessoaBrain @pessoabrain.bsky.social · 09/09/2026
Paper spearheaded by Paul Taylor and Gang Chen from the AFNI team with lots of others. I had a very minor role here but highly recommend the thinking behind the idea. #neuroskyence
272
Reposted by Jeremy Lefort-Besnard
Gaël Varoquaux @gaelvaroquaux.bsky.social · 08/09/2026
Nice post, as always. For parallelism in Python, I almost always use Joblib joblib.readthedocs.io/en/stable: not only does it tend to simplify code, but also it pulls many tricks for you (auto-batching, or reusing a process pool...)
joblib.readthedocs.io
Joblib documentation — joblib 1.6.0 documentation
223
Reposted by Jeremy Lefort-Besnard
Russ Poldrack @russpoldrack.org · 08/09/2026
A brief introduction to parallelism and multithreading russpoldrack.substack.com/p/a-brief-in... - the latest in my Better Code, Better Science series
russpoldrack.substack.com
A brief introduction to parallelism and multithreading
Better Code, Better Science: Chapter 10, Part 8
171
Reposted by Jeremy Lefort-Besnard
Richard McElreath 🐈‍⬛ @rmcelreath.bsky.social · 06/09/2026
Here's a projection playground, you can set the center anywhere you like. please don't break the planet kuboon.github.io/equal-earth/
a very weird map projection with the north pole at the center
35015
Reposted by Jeremy Lefort-Besnard
Berna Devezer @devezer.bsky.social · 04/09/2026
i think i've become so cynical by now that this was exactly what i'd have expected to see after seeing the ignobel discourse. fascinating how credulous we still are about this sort of thing. both regarding results that confirm our expectations and ignobel stuff in general.
2204
Reposted by Jeremy Lefort-Besnard
Paul Taylor @afni-pt.bsky.social · 31/08/2026
Have you ever felt like something was missing from your #fmri results? Have you been worried about potential reproducibility issues? Well, this new @natmethods.nature.com article with 40+ coauthors shows how the simple change to transparent thresholding helps both. Go Figure! rdcu.be/B4tikBETdQZa
A comparison of old and new thresholding approaches for fMRI data. The newer, transparent thresholding allows for interpreting context, more complete modeling, a reduction of biases and an increase in meaningful content. This all leads to more accurate interpretations and improved reproducibility evaluations.
19151
Reposted by Jeremy Lefort-Besnard
Mel🦄 @enbyandbi.eurosky.social · 30/08/2026
On rappelle que, même si les IA nous rendaient plus intelligent·es, le rapport bénéfice/risque serait encore trop élevé dans un monde qui crame. Mais, au minimum, elle nous enlève des savoir-faire.
210543
Reposted by Jeremy Lefort-Besnard
Berna Devezer @devezer.bsky.social · 28/08/2026
On this above note, also see
1231
Reposted by Jeremy Lefort-Besnard
Konrad Kording @kordinglab.bsky.social · 25/08/2026
Thread. I have yet to see a proper explanation of how LLMs actually work that is targeted to an informed general audience. So here is an explainer that I put together along with Tim @countzerozzz (inspired by @CIFAR_News LMB discussions).
1723160
Reposted by Jeremy Lefort-Besnard
Konrad Kording @kordinglab.bsky.social · 27/08/2026
Using AI in neuroscience labs. Great article at @thetransmitter.bsky.social by @nicoschuck.bsky.social. www.thetransmitter.org/artificial-i...
thetransmitter.org
How, when and why to use agentic AI in our neuroscience labs
With agentic AI’s rapid infiltration into science, researchers need to develop AI use policies and practices.
0187
Reposted by Jeremy Lefort-Besnard
Dirk Gütlin @gutlin.bsky.social · 25/08/2026
You might find it funny or relievedly turn back to scream at ChatGPT in all caps, but this is a valid question and an interesting read. 🤷‍♀️😁 Large language models do not have emotions www.nature.com/articles/s41...
nature.com
Large language models do not have emotions - Nature Human Behaviour
Nature Human Behaviour - Large language models do not have emotions
054
Reposted by Jeremy Lefort-Besnard
Ida Momennejad @neuroai.bsky.social · 19/08/2026
A crucial part of doing science is the art of expressing & engaging with disagreement. Cheng et al show 11 AI models are sycophantic & make users more convinced they're right & less willing to take responsibility or repair conflict. Q. As we adopt AI in science how can we preserve healthy discourse?
science.org
Sycophantic AI decreases prosocial intentions and promotes dependence
Despite rising concerns about sycophancy—excessive agreement or flattery from artificial intelligence (AI) systems—little is known about its prevalence or consequences. We show that sycophancy is wide...
2254
Reposted by Jeremy Lefort-Besnard
Joseph Cox @josephcox.bsky.social · 17/08/2026
New from 404 Media: we solved which AI company is buying massive shipments of rare books, scanning and destroying them to train AI. We put an Apple AirTag in a rare book, followed it. It ended up at an Amazon facility. Its logo is a dinosaur ripping through a book www.404media.co/we-tracked-a...
404media.co
We Tracked a Shipment of Rare Books. It Ended at an Amazon AI Training Facility
We placed a tracking device in a shipment of rare books to see which AI company was buying it, and found an Amazon facility where Amazon scans and destroys books.
12537942368
Reposted by Jeremy Lefort-Besnard
Hilary Mason @hilarymason.com · 19/08/2026
This thread helped me understand that the outrage about Anthropic's watermarking is the confluence of a lack of understanding of how LLMs (or any statistical language model) works and general angst at the current economics.
191
Reposted by Jeremy Lefort-Besnard
Ash Davies 🇩🇪 🏴󠁧󠁢󠁷󠁬󠁳󠁿 @ashdavies.dev · 13/08/2026
> I had a conversation ... about how a developer's job is now is to teach the AI to be better, make fewer mistakes, etc. To review the code. And to test the changes that the AI makes. This is a horrible existence This is so relatable. brettcodes.com/im-done-usin...
brettcodes.com
I'm done using AI
Why I'm stopping using AI for coding (and anything else) after using it earnestly for a year and coming to understand the harms it causes on personal, societal, and environmental levels.
3135
Reposted by Jeremy Lefort-Besnard
Russ Poldrack @russpoldrack.org · 27/07/2026
Twelve quick tips for AI-assisted coding in science journals.plos.org/ploscompbiol... - our latest by @ericwbridgeford.bsky.social with @thisisadax.bsky.social , Zijiao Chen, Zhicheng Lin, @hritz.bsky.social and @joachim.cidlab.com
journals.plos.org
Twelve quick tips for AI-assisted coding in science
While AI coding tools have demonstrated potential to accelerate software development, their use in scientific computing raises critical questions about code quality and scientific validity. In this pa...
14010
Reposted by Jeremy Lefort-Besnard
Richard McElreath 🐈‍⬛ @rmcelreath.bsky.social · 16/07/2026
This looks useful. A typical meta-analysis in ecology mixes experiments with observational studies, mixes coefficients with different controls and therefore different causal meanings. I would not normally say "hey the folks in medicine are doing it right, imitate them" but in this case maybe
25115
Reposted by Jeremy Lefort-Besnard
Russ Poldrack @russpoldrack.org · 07/07/2026
Better Code, Better Science is back after a bit of travel. We also have a new web site (bettercode-book.org). Next week I will post a number of bonus articles based on sections added since the original posts - stay tuned! russpoldrack.substack.com/p/statistica...
russpoldrack.substack.com
Statistical calibration using randomization
Better Code, Better Science: Chapter 9, Part 6
1182
Reposted by Jeremy Lefort-Besnard
Richard McElreath 🐈‍⬛ @rmcelreath.bsky.social · 02/07/2026
I think this kind of thing deserves a response that reveals the assumptions that make "Rabois World" correct, including no mediators or colliders. I don't have time for that though. So here's my old series on what's wrong with thinking like a regression: elevanth.org/blog/2021/06...
elevanth.org
Regression, Fire, and Dangerous Things (1/3)
It isn't my job to disappoint people, but I'm good at it.
1395
Reposted by Jeremy Lefort-Besnard
Gaël Varoquaux @gaelvaroquaux.bsky.social · 02/07/2026
A fun podcast with Maxime Gabella about our dreams about AI. The next frontier of AI is building agents that use statistical learning to tackle problem-specific challenges from data. Agents come with promises, but they need a statistical harness, as @probabl.ai's www.youtube.com/watch?v=cNpv...
youtube.com
Can AI Become a Real Data Scientist? | Gaël Varoquaux on scikit-learn, Probabl & Scientific Judgment
YouTube video by Maxime Gabella
0104
Reposted by Jeremy Lefort-Besnard
Dr. Serge Zaka @sergezaka.bsky.social · 02/07/2026
Cette animation est saisissante. Elle montre, en seulement quelques secondes, l'accélération spectaculaire de la fréquence des vagues de chaleur au cours des 126 dernières années en France. 1/3
13863475
Reposted by Jeremy Lefort-Besnard
Dr. Serge Zaka @sergezaka.bsky.social · 26/06/2026
+1000% de mortalité chez les poules, +200% chez les porcs, +45% chez les bovins. Plusieurs millions d'animaux d'élevage sont morts sous l'effet de la canicule. 1/13
411204992
Reposted by Jeremy Lefort-Besnard
Protocol Labs @protocollabs.bsky.social · 25/06/2026
Konrad Kording argues that the best way to understand the brain may be to read its wiring directly rather than infer it from the outside. @kordinglab.bsky.social joins @juanbenet.bsky.social for Ep 5 of the Juan Benet Podcast. Full episode: youtu.be/FHQfmJEpRmU
youtu.be
The Power of a Single Neuron and a Path to Simulating the Brain | Dr. Konrad Kording
YouTube video by Juan Benet Podcast
1136
Reposted by Jeremy Lefort-Besnard
Russ Poldrack @russpoldrack.org · 29/06/2026
“agents should not directly touch the text of the paper. Paper text should be written by a human.” This is my lab’s explicit policy as well. As always, @mcxfrank.bsky.social nails it.
2457
Reposted by Jeremy Lefort-Besnard
Gaël Varoquaux @gaelvaroquaux.bsky.social · 29/06/2026
🧑‍💻🧑‍🏫 I'm recruiting a post-doc to work on Tabular Foundation Models, one of the hotest topics in AI, where we are at the leading edge team.inria.fr/soda/files/2... This is an opportunity to develop the next-level tabular AI, blending deep learning and tables.
team.inria.fr
33320
Reposted by Jeremy Lefort-Besnard
Thomas Yeo @bttyeo.bsky.social · 25/05/2026
For those who are attending @ohbmofficial.bsky.social brainhack, I will be giving a short presentation of this work at the neuroimaging statistics workshop, co-hosted with brainhack: sites.google.com/view/nsw2026
sites.google.com
NeuroStats2026
02314
Reposted by Jeremy Lefort-Besnard
Paul Taylor @afni-pt.bsky.social · 09/06/2026
Job ad (1/2): Postdoc fellowship in the AFNI group (SSCC, NIMH, NIH, USA). FMRI/MRI methods dev, software design, statistics, analysis, data viz, quality control, machine learning and more! Please see: discuss.afni.nimh.nih.gov/t/job-ad-pos... I am happy to chat at OHBM+Brainhack. Please share!
22625
Reposted by Jeremy Lefort-Besnard
Françoise Halper @frhalper.bsky.social · 10/01/2026
La start-up française #MistralAI signe un partenariat stratégique avec le ministère des #Armées 👉"Garantir la #souveraineté et l’excellence technologique de la défense française" 👉La mise en œuvre du contrat sera supervisée par l’AMIAD www.bfmtv.com/tech/intelli...
042
Reposted by Jeremy Lefort-Besnard
Ida Momennejad @neuroai.bsky.social · 03/01/2026
"All intelligence is collective intelligence: each of us consists of a huge number of cells working together to generate a coherent cognitive being with goals, preferences, & memories that belong to the whole and not to its parts" @drmichaellevin.bsky.social paper pubmed.ncbi.nlm.nih.gov/37204591/
2375
Reposted by Jeremy Lefort-Besnard
Sebastian Raschka (rasbt) @rasbt.bsky.social · 28/12/2025
I think of it as this: LLMs lower the barrier of entry, and they make coders (beginners and experts) more productive. It's still worth investing in becoming an expert, because then you will get even more out of LLMs and will be able to deliver even better results.
4333
Reposted by Jeremy Lefort-Besnard
Adam Bonica @adambonica.bsky.social · 06/12/2025
Study after study shows campaign ads barely move the needle. So where does money’s real power come from? I ranked the five ways money corrupts politics—from least to most corrosive. What I’ve learned from 15 years of tracking political money:
open.substack.com
Money Doesn't Buy Elections. It Does Something Worse.
Campaign ads barely move the needle. The real influence is hiding in plain sight.
461217541
Reposted by Jeremy Lefort-Besnard
Sebastian Raschka (rasbt) @rasbt.bsky.social · 23/11/2025
Lots of interesting LLM releases last week. My fav was actually Olmo 3 (I love the Olmo series due to their full open-sourceness and transparency). If you are interested in reading through the architecture details, I coded it from scratch here: github.com/rasbt/LLMs-f...
07310
Reposted by Jeremy Lefort-Besnard
Konrad Kording @kordinglab.bsky.social · 23/11/2025
Interesting mystery. It is known that animals can learn to control neurons pretty much anywhere in the brain. But they can not learn to ignore hunger which probably means they can't turn of hunger sensing neurons. How is that avoided in the brain? They even have DA inputs.
14212
Reposted by Jeremy Lefort-Besnard
Johanna Joyce @johannajoyce.bsky.social · 05/11/2025
New issue of Nature - with NINE studies on #brain #development from the BRAIN Initiative Cell Atlas Network (BICAN) 🧠🧪🔬 An amazing set of resources for all scientists working on the brain! 🧠 Immersive feature: www.nature.com/immersive/d4... 🧠 Perspective: www.nature.com/articles/s41...
Cover of this week's Nature showing a brain rendering

Cover caption from the journal:
Brain development:
Our ability to process information into complex emotions, behaviours and decisions relies on the rich diversity of cell types that make up the human brain. Uncovering the molecular and cellular events that take place during brain development could reveal not only the mechanisms that give rise to this diversity but also shed light on how this process might go awry in neurodevelopmental disorders such as autism and schizophrenia. In this week’s issue, the BRAIN Initiative Cell Atlas Network (BICAN) builds on its previous work creating atlases of cell types in the adult mouse, non-human primate (NHP) and human brains to present cell-type atlases of the developing human, mouse and NHP brains. Across a suite of papers, nine of which are published in Nature, the researchers uncover the complex programs through which cell types emerge during brain development in humans and animals, revealing both the shared and unique features of the human brain. The latest work, along with future research directions, is summed up in a Perspective article by Tomasz Nowakowski and colleagues
09140
Reposted by Jeremy Lefort-Besnard
Berna Devezer @devezer.bsky.social · 24/10/2025
I couldn't be more on board with this. Focusing on corruption/fraud/misconduct at the level of individual scientist appears to be misguided and counterproductive when systemic forces and selection biases are at play. This appears to be true even beyond the context of industry manipulation.
3665
Reposted by Jeremy Lefort-Besnard
Empenn research team @empenn-lab.bsky.social · 18/09/2025
🚀 We’re hiring! Are you curious about how young children's brains develop? 🧠 👉 buff.ly/09zJXX2 💻 Postdoc (Computer Science) - on brain atlases & developmental trajectories in young kids buff.ly/sDffS1b 📋 Project Coordinator - lead the implementation & monitoring buff.ly/tMQ2gZn
012
Reposted by Jeremy Lefort-Besnard
Réseau français de la recherche reproductible @reproductible-fr.bsky.social · 16/10/2025
Dans un contexte où la #Reproductibilité est clé pour la crédibilité de la science, ces journées rassemblent des scientifiques pour tenter de reproduire des études publiées. 👏 Bravo à toutes et tous ! 👉 Lire l'article : www.lemonde.fr/sciences/art...
lemonde.fr
Les « jeux de la réplication », ou quand la science s’amuse à se reproduire
Une vingtaine de chercheurs ont traqué, le 3 octobre, les failles de plusieurs articles de sciences sociales parus dans des revues de renom.
155
Reposted by Jeremy Lefort-Besnard
Aperture Neuro @apertureohbm.bsky.social · 16/10/2025
Lefort-Besnard and Pron et al. present a survey and review on current infrastructures for sharing human neuroimaging data in the EU: doi.org/10.52294/001... @cmaumet.bsky.social @ohbmossig.bsky.social #OpenDatasets #SpecialIssue
134
Jeremy Lefort-Besnard @jlefortbesnard.bsky.social · 16/10/2025
Vous êtes en dernière année d’école d’ingénieur ou de master spécialisé IA/embarqué/robotique et intéressé(e) par les problématiques autour de la distillation et de la quantization de modèles et des techniques d'optimisation de l'inférence ? Ce stage est pour vous. www.linkedin.com/posts/activi...
linkedin.com
🚀 Offre de stage – avec Daniel Trezentos, nous proposons un stage de fin d'études au sein de l'AMIAD - Agence Ministérielle pour l'IA de Défense sur la thématique de la compression et l&#3...
🚀 Offre de stage – avec Daniel Trezentos, nous proposons un stage de fin d'études au sein de l'AMIAD - Agence Ministérielle pour l'IA de Défense sur la thématique de la compression et l'optimisation d...
001
Jeremy Lefort-Besnard @jlefortbesnard.bsky.social · 04/10/2025
L’électronique ou la robotique t’intéresse ? 🤖 ➡️ RDV le mardi soir 18h-20h chaque semaine à l’ #Edulab de #Rennes 📍 Viens aussi découvrir aujourd’hui pendant l’ #OpenLab 14-18h (même endroit)
000
Reposted by Jeremy Lefort-Besnard
Dr Cyril Pernet @cyrilrpernet.bsky.social · 17/03/2025
A sensitivity analysis of preprocessing pipelines: toward a solution for multiverse analyses with @melanieganzben1.bsky.social et al. Basically, adapting meta-analysis tools accountting for the correlations between pipelines direct.mit.edu/imag/article...
Difference in SERT availability across pipelines using different multiverse estimators
0199
Reposted by Jeremy Lefort-Besnard
Imaging Neuroscience @imagingneurosci.bsky.social · 01/05/2025
New paper in Imaging Neuroscience by Elodie Germani, Camille Maumet, et al: On the validity of fMRI mega-analyses using data processed with different pipelines doi.org/10.1162/imag...
0128
Reposted by Jeremy Lefort-Besnard
Thomas Yeo @bttyeo.bsky.social · 17/07/2025
1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...
317182
Reposted by Jeremy Lefort-Besnard
Camille Maumet @cmaumet.bsky.social · 11/02/2025
🤔 You’d like to share your research code but not sure how to…? 🔓*Very* excited to finally share this work led by @nicolasrougier.bsky.social in which we propose a practical set of guidelines for opening code. inria.hal.science/hal-04930405...
An infographic providing an overview of the CODE roadmap for sharing research software and code: Open Document Execute Collaborate.
174
Reposted by Jeremy Lefort-Besnard
Ouvrir la science @ouvrirlascience.bsky.social · 14/04/2025
Testez la version bêta du catalogue des logiciels libres de la recherche académique ! Une enquête est mise à disposition pour consolider et faire évoluer cette toute première version. www.ouvrirlascience.fr/lancement-de... #logiciellibre #opensource #scienceouverte #openscience
02213
Reposted by Jeremy Lefort-Besnard
Imaging Neuroscience @imagingneurosci.bsky.social · 17/04/2025
On this day 2y ago we announced our move away from the huge profits that NeuroImage makes for Elsevier. Things at Imaging Neuroscience have progressed fantastically thanks to the support for this move by the brain imaging community and @mitpress.bsky.social. 565 papers published already!
Screenshot of the original tweet announcing our move, with 5K likes and 2M views.
020161