Just published in Science (!!), and a landmark achievement for ARISE.

An LLM (o1-preview) outperformed physicians on multiple clinical reasoning tasks, including blinded second opinions on real emergency department cases from BIDMC.

Also notable: physician raters were unable to tell whether a differential came from AI or a human. One rater responded “can’t tell” in 83.6% of cases, the other in 94.4%.

Most likely near-term use case: AI as a high-quality second-opinion and reasoning support tool, especially early in the diagnostic process when physicians have incomplete information.

Peter Brodeur, MD, MA, Thomas Buckley, Adam Rodman, Jonathan H. Chen, Arjun Manrai

With a fantastic team: Robert Gallo, Zahir Kanjee, Evelyn Bin Liang, Priyank Jain, Stephanie Cabral, Raja-Elie Abdulnour, Adrian Haimovich, Andrew Olson, Daniel Morgan, Haadi Mombini, Liam McCoy, Christopher Lucas, Jason Hom, Jason Freed, MD, Daniel Restrepo, Eric Horvitz

+ would not have been possible without Yevgeniya Nusinovich's guidance and support throughout entire process. | 54 comments on LinkedIn

2031. Vous êtes consultant senior depuis 15 ans. L'IA fait votre recherche, vos slides, vos premiers jets. Vous validez. Depuis 5 ans, vous n'avez pas rédigé un paragraphe seul. Un matin, le système tombe. Vous ouvrez un document vierge. Rien ne vient. 🧠

L'étude du MIT Media Lab l'a montré. 54 adultes suivis pendant 4 mois. Connectivité neuronale réduite de 55% chez les utilisateurs de LLM. Time Magazine l'a titré. Le Data Science Collective a parlé de "47% Collapse in Brain Activity".

Ce qui est en train de mourir :
👉 Le muscle de la rédaction. 83% des utilisateurs de LLM sont incapables de citer un passage du texte qu'ils viennent d'écrire. Ils ne reconnaissent pas leur propre travail.
👉 La connectivité même après arrêt. Quand les utilisateurs habituels reprenaient une tâche sans assistance, leur connectivité restait réduite. Le cerveau peinait à redémarrer seul.
👉 La capacité à détecter les erreurs. Harvard et BCG : +40% de performance dans la zone de compétence de l'IA. Moins 19% en dehors. Le danger n'est pas que l'IA se trompe. C'est que vous ne sachiez plus repérer où.

Trois antidotes :
1️⃣ Faire avec, pas faire après. Si vous n'intervenez qu'à la fin pour valider, vous êtes un dirigeant qui lit un rapport sans avoir participé à aucune réunion. Réintroduisez des checkpoints humains.
2️⃣ Orchestrer, pas subir. Construire le système, définir les étapes, choisir où intervenir. La première posture construit de l'expertise. La seconde accumule de la dette cognitive.
3️⃣ Garder une discipline de pensée autonome. Régulièrement, fermez l'IA. Rédigez seul. C'est la salle de musculation du jugement. Par hygiène cognitive.

Ma conviction : la dette cognitive est le danger silencieux de l'ère agentique. Invisible au début. Catastrophique quand on découvre qu'on ne sait plus évaluer si les résultats sont bons.

Série "La Mort du Conseil" [6/10]. Le muscle intellectuel est mort. Demain : la confiance passe sur la table.

🚀 Dirigeants : votre consultant valide-t-il encore ce qu'il produit, ou appuie-t-il sur un bouton ? 👉 https://lnkd.in/e6k46944
🎓 Consultants : -55% de connectivité neuronale. Votre cerveau est votre outil. Protégez-le. 👉 https://lnkd.in/eaJd3bZ8
🎯 Masterclass gratuite : de consultant à architecte IA en 1h. 👉 https://lnkd.in/eZGGrvvY
🚀 Un projet IA ? Un séminaire CODIR ? Un bootcamp interne 👉 https://decisionia.com/rdv

-55% de connectivité neuronale. 83% incapables de citer leur propre texte. Et vous, vous savez encore ce que vous avez écrit ce matin ? 👇 | 79 comments on LinkedIn

🎓 10 Educational Leadership Mistakes That Kill Teacher Productivity — And How I Would Fix Them

Most schools don’t struggle because teachers lack talent.

They struggle because leadership systems reduce teacher productivity.

When teachers lose:
❌ time
❌ clarity
❌ motivation
❌ ownership

students eventually pay the price.

These are the 10 leadership mistakes I frequently observe in schools — and how I would address them as a principal. 👇

1️⃣ Unclear Priorities
👉 Too many tasks, little focus
✔ Fix: Define 3 clear school priorities and align work accordingly

2️⃣ Excessive Administrative Load
👉 Teachers spend more time on paperwork than planning
✔ Fix: Simplify and digitize routine processes

3️⃣ Last-Minute Communication
👉 Sudden updates create stress and poor execution
✔ Fix: Weekly communication system with advance planning

4️⃣ Meetings Without Outcomes
👉 Long discussions, no decisions
✔ Fix: Agenda-based, time-bound meetings with action points

5️⃣ Micromanagement
👉 Constant checking kills ownership
✔ Fix: Trust + accountability + coaching culture

6️⃣ Slow Decision-Making
👉 Small issues remain pending too long
✔ Fix: Fast escalation and quick resolution systems

7️⃣ Lack of Recognition
👉 Consistent effort goes unnoticed
✔ Fix: Monthly appreciation and recognition culture

8️⃣ Fear-Based Leadership
👉 Teachers stop innovating when fear increases
✔ Fix: Safe feedback and learning environment

9️⃣ Weak Collaboration
👉 Departments work in silos
✔ Fix: Cross-team planning and peer learning systems

🔟 Ignoring Teacher Well-Being
👉 Burnout silently reduces productivity
✔ Fix: Balanced workload and supportive leadership

🎯 My Leadership Belief

If I become principal, my first goal would not be control.

It would be removing the barriers that stop teachers from doing their best work.

Because when teachers thrive:
✔ students perform better
✔ culture improves
✔ parents trust more
✔ results become sustainable

A strong principal doesn’t just manage operations.

A strong principal multiplies teacher productivity.

— Mohini Sudarshan Bedge

#SchoolLeadership
#PrincipalLeadership
#EducationLeadership
#SchoolImprovement
#TeacherProductivity
#EducationalManagement
#innovation
#cbseschool

💡 Les IA ne s'étonnent pas (encore)

Le philosophe Charles S. Peirce appelait abduction « la seule opération logique qui introduise la moindre idée nouvelle ». Face à un fait surprenant, elle propose une explication possible et franchit ce qu'il nomme un seuil vertical, c'est-à-dire qu'elle change de niveau explicatif.

Or un modèle de langage fonctionne par extrapolation horizontale. Sa pente est statistique. Il prolonge ce que son corpus a déjà dit mais ne franchit pas de seuil.

L'analyse stratégique vit pourtant de ce franchissement. C'est même ce qui la distingue d'un simple résumé documenté. Elle prend de la valeur au moment où un concept né dans un autre champ rend soudain lisible une situation qui semblait inerte, ou quand une page d'un philosophe d'autrefois éclaire un blocage organisationnel d'aujourd'hui. Ces rapprochements ne sortent d'aucune statistique de corpus. Ils supposent quelqu'un qui sache dans quelle direction creuser, et pourquoi cette direction.

Peirce parlait d'imagination et de créativité à l'œuvre dans l'abduction. Mais imagination et créativité ne sortent pas du néant. Elles puisent dans un capital sédimenté qui ne s'acquiert qu'avec patience. Voilà ce que la culture du retour sur investissement immédiat dévalorise depuis longtemps : la lecture longue, la connaissance qui dort, les détours dont l'utilité ne se révèle parfois que des années après.

Le généraliste cultivé est cet investisseur patient. Il accumule des connaissances dont il ne peut pas prévoir l'usage, en sachant qu'une partie restera dormante et qu'une autre, le jour venu, vaudra son pesant d'or. À mesure que les LLM externalisent la spécialité, cette posture redevient stratégique. Le spécialiste pur est rattrapé par les modèles, et celui qui se contente de compiler aussi. Ce qui reste rare, c'est celui qui sait quel champ aller activer pour faire surgir un rapprochement surprenant, fécond.

L'abduction a ses défauts, et ma thèse en pointe trois : la tentation des hypothèses séduisantes mais fausses, la prolifération de pistes concurrentes qu'on ne peut pas trancher rapidement, et le biais de confirmation qui s'amplifie quand l'analyste s'attache à son intuition initiale. C'est pourquoi le rapprochement fécond ne suffit pas. Sans épreuve des hypothèses, sans recherche de ce qui les infirmerait, l'abduction reste un jeu d'esprit. Ce qui distingue l'analyste du conférencier brillant, c'est précisément qu'il sait conduire les deux temps.

Et vous, quel auteur fréquentez-vous en ce moment sans savoir où ça va vous mener ?

(Extrait de ma thèse de doctorat, 2025 - Lien en commentaire 👇)

#VeilleStratégique #IntelligenceÉconomique #PenséeCritique #Abduction #IA

NATURE | The human proteome just got bigger — and stranger.

Nature: https://lnkd.in/eGdXNWFb
News: https://lnkd.in/e2pHyFCn

A landmark study published this week in Nature by the TransCODE Consortium analyzed 95,520 proteomics experiments and found that roughly 25% of ~7,300 previously "non-coding" sequences actually produce detectable proteins. These tiny molecules, called microproteins, have been hiding in parts of the genome we long assumed were silent.

But the most exciting concept to emerge from this work is the "peptidein" — a new category of molecule defined as a confirmed translation product that doesn't yet meet the bar for a conventional protein-coding gene. Not quite a protein. Not nothing. A whole new class sitting in between, waiting to be understood.

The team also:
→ Developed ORBL, a new evolutionary analysis tool that measures how conserved the structural "openness" of a reading frame is across species — independent of amino acid sequence
→ Identified one peptidein from the OLMALINC RNA that turns out to be essential for cell survival in 85% of tested cancer cell lines
→ Formally annotated new protein-coding genes in databases like GENCODE and PeptideAtlas

This matters well beyond basic science. Microproteins and peptideins are showing up in cancer immunopeptidomics data — meaning they may be visible to the immune system and could be targetable by T-cell therapies. The field of cancer immunotherapy just got a new list of leads to chase.

We're only beginning to understand what the genome actually encodes. Papers like this remind us how much biology is still waiting to be found.

#Proteomics #Genomics #CancerResearch #Immunotherapy #Biochemistry #Science

Picture: Nature

Very much looking forward to delivering the keynote for this event. Thanks Vikki Hill for the kind invitation. Looking forward to meeting your colleagues and having rich discussions about joyful learning and teaching!

Everyone's rushing to deploy AI.

Almost no one is asking whether their organisational knowledge is actually ready for it.

A large language model will reflect the quality of your knowledge ecosystem. Feed it outdated policies, inconsistent documentation, and tribal knowledge that lives only in people's heads — and you'll get confident, fluent, wrong answers. At scale.

AI doesn't fix broken knowledge management. It exposes it.

For most of my career, I've had some version of the same conversation. "Yes, knowledge management matters." [Nods around the table.] "We really should invest in it properly." [More nodding.] Then the budget cycle comes around and KM is once again the first thing cut.

AI is changing that — not because it solves the knowledge problem, but because it makes the consequences of ignoring it impossible to dismiss.
I've written a longer piece on why I think AI might be the best thing that ever happened to knowledge management — and what organisations need to get right before they find out the hard way.

Would love to know what you're seeing on the ground. 👇

#knowledgemanagement #innovation #AI #sustainability| 33 commentaires sur LinkedIn

👏 27,3% des fonds marins ont été cartographiés.
Alors qu’on a cartographié toute la planète Mars 🤔

L’océan reste une Terra Incognita qui doit susciter humilité et soif de connaissances.

En 2025, 4 millions de km² ont été cartographiés : l’équivalent de l’Union Européenne en un an🤯

L’objectif est d’avoir cartographié l’ensemble du fond des océans d’ici 2030.

🧑‍🔬 Cela ouvrira encore bien d’autres champs de recherche : biodiversité des abysses, prédiction des tsunamis, étude des sources hydrothermales…

🗺️ #30daysmapchallenge, jour 36/45 : participez en commentaire avec vos idées 👇


__
❤️ Retrouvez moi sur Instagram pour (re)découvrir le monde maritime : https://lnkd.in/dHDE78kB| 11 commentaires sur LinkedIn

"Can you remain active in wet-lab research as a biomedical educator?" This was the question I tackled during the Department of Pharmacology | NUS Medicine's academic staff retreat.

Surrounded by the goodwill and positive energy of longtime colleagues and friends, I was able to share my personal journey honestly, both achievements and failures, fulfilment and dejection.

My take homes for the session:
1. Prioritize: You may have loved bench-side experiments as a PhD student / post-doc, and may even have excelled in it, but as a biomedical educator, wet-lab research is no longer your primary KPI. Spend the bulk of your efforts getting good on reflective teaching, curriculum development / revamp, education outreach, pedagogical research & service... THEN, if you still have the bandwidth, go for it (see 2).

2. Understand Your Motivation: It is 6pm, you are tired from a whole day of teaching, meetings and exam marking. Do you still want to analyze that cell-culture data, write that paper, take that Zoom meeting with your students? Why? What's in it for you now that the work won't (directly) help with your career advancement?

3. Develop Expertise: I started my PhD working on dementia neurochemistry, then at some point fell into blood-based biomarkers (BBMs). Through opportunities and generous support (see 6), I began to develop a real interest in this fast-moving field, and published fairly regularly on it. Gradually, I got invited to lecture on BBMs, and to participate in programmatic grant proposals in which BBM is a component. So, take the time (multiple years in my case) to build expertise and be a go-to person for an area in which you have strong interest (see 4).

4. "Give up" and Collaborate: A colleague once told me their paper from a C(ell) N(ature) S(cience)-level journal took multiple years to publish (it was >1.5 years just to do the requested experiments for a revision). Given the high bar set by top multidisciplinary journals nowadays, it is difficult enough to lead all aspects of the research (including procuring resources) for such papers as a tenure-track PI, and virtually impossible (IMHO) to do so as an educator. So I focus on being a good collaborator / theme lead and contributing to worthwhile, impactful research programs needing my area of expertise.

5. Make Your Research Edu-Relevant: At every opportunity, translate your research into your teaching. I have given lectures and seminars on BBMs to diverse audiences, from secondary school students through to practicing doctors, a deeply rewarding experience!

6. Be Thankful: "It takes a village..." to nurture a biomedical researcher. I certainly have my share of mentors who generously guided, taught and enabled me, collaborators who enthusiastically shared their expertise and data with me, and students / team members who steadfastly supported me through peaks and valleys. To all these wonderful people I say a loud "thank you" for enriching my professional life so abundantly!

En 2007, Apple lance l'iPhone. 
Et en 2009, la fécondité mondiale décroche.

Pendant longtemps, les démographes ont cru à une oscillation passagère. Précisément parce qu’avant 2009, la fécondité baissait, remontait, baissait. Alors on s’est dit qu’il n’y avait rien d'alarmant, que les couples faisaient des enfants plus tard, pas qu’il en faisaient moins.

Sauf que, 15 ans plus tard, il a bien fallu se rendre à l'évidence : États-Unis, Mexique, Italie, Argentine, Iran, Tunisie, Corée, France, la fécondité n'est jamais remontée, et pour la première fois dans l'histoire moderne, l’effondrement est synchrone et planétaire.

Ce qui est planétaire aussi, c'est ce petit rectangle lumineux glissé dans la poche de 4 milliards d'êtres humains. Car oui, je crois profondément que le smartphone et les réseaux sociaux ont détruit la rencontre. Ils nous donnent l'illusion d'une proximité relationnelle qui est en réalité superficielle. Une bulle de notifications, de stories et de likes qui simule la présence des autres tout en nous laissant assis seul sur notre canapé. On pense « voir » ses amis parce qu'on défile dans leur quotidien sur Instagram. On pense « parler » parce qu'on s'envoie des memes. Mais on ne s'embrasse plus, on ne se dispute plus, on ne se touche plus 😕

A tel point qu’aux États-Unis, la part des adolescents qui voient leurs amis en chair et en os moins d'une fois par mois est passée de 3 % entre 1990 et 2010 à 10 % en 2019, pendant que 46 % d’entre eux se déclarent « constamment en ligne » d’après le Financial Times.

Or, c'est précisément cette génération-là qui devrait, aujourd'hui, tomber amoureuse. Sauf qu'on ne tombe amoureux de personne quand on ne rencontre personne…

Au fond, je crois que le smartphone n'a pas tué le désir d'enfant. Il a tué l'écosystème dans lequel ce désir naît. La fête improvisée, la conversation qui dérive jusqu'à 3h du matin, l'ennui partagé qui pousse deux individus à inventer quelque chose ensemble. Le smartphone a supprimé le temps mort. Et, avec lui, l'élan instinctif qui pousse à se tourner vers la personne assise à côté 👋

Avant d'être une crise de berceaux, la dénatalité est donc avant tout une crise de canapés. Parce qu’il n'y a tout simplement plus deux personnes dessus. | 69 comments on LinkedIn

One concept worth understanding in evidence-based work is the Woozle Effect.

It occurs when a claim gains the appearance of strong evidential support through repetition and citation, even though the original research did not clearly establish it.

- A preliminary or limited finding is published.
- Later summaries describe it with slightly more certainty.
- Secondary sources cite those summaries rather than the primary paper.
- Caveats, exclusions and uncertainty gradually disappear.
- Repetition creates familiarity.
- Familiarity is mistaken for reliability.
- The proposition becomes treated as established knowledge.

Eventually, the claim can feel settled simply because it has circulated widely enough.

The Woozle Effect is usually driven less by dishonesty than by repetition, assumption and uncritical citation. It often reflects ordinary human tendencies: confirmation bias, inherited beliefs, deference to consensus and failure to revisit primary sources.

The result is that non-facts can become common sense. In any field that relies on research, policy or expert opinion, that carries obvious risk.

The Woozle Effect is not merely about misinformation. It is about how intelligent people, acting in good faith, inherit certainty they did not build.

That is why disciplined analysis often begins with a deceptively simple question:

What did the original evidence actually show?

Repetition can amplify a claim but it cannot validate one. | 23 comments on LinkedIn

What feels obvious to you… might not be obvious to someone else.

That gap shows up more often than we think.

Our latest blog explores the 𝗖𝘂𝗿𝘀𝗲 𝗼𝗳 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 and why experts often struggle to explain things clearly. Once we understand something deeply, it’s hard to remember what it felt like not to.

So we skip steps.
We use language that feels natural to us.
We move faster than others can follow.

And that’s where learning breaks.

In learning design, expertise isn’t the goal. Clarity is.

That means slowing things down.
Breaking ideas into steps.
Saying things simply, even when they feel obvious.

Because what’s clear to you isn’t always clear to someone else.

And the moment it becomes clear… that’s when learning actually happens.

📌 Write to 𝗲𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴@𝗹𝗲𝗮𝗿𝗻𝗻𝗼𝘃𝗮𝘁𝗼𝗿𝘀.𝗰𝗼𝗺 to craft learning that transforms behaviour.

#LearningDesign #LearningScience #WorkplaceLearning #InstructionalDesign

https://lnkd.in/g98_6pha

Spread the loveIntroduction to Social Media Mining in Health The advent of social media has transformed how we communicate, share information, and engage with health-related topics. As we delve into the years 2015 to 2025, the phenomenon known as social media mining in health emerges as a critical...

#PositiveAcademia #288: This one is for academics in the later stages of their careers in particular: an article to help you reflect about how you want to shape your career and life after 20+ years in academia.

We recently discussed transitioning to retirement in our CYGNA Senior meeting (see: https://lnkd.in/eKmP3mN5) and this article came up. It is a fascinating discussion about the difference between fluid and crystallised intelligence that is highly relevant for academic careers.

We all agreed we'd rather be Bach than Darwin:
"When Darwin fell behind as an innovator, he became despondent and depressed; his life ended in sad inactivity. When Bach fell behind, he reinvented himself as a master instructor. He died beloved, fulfilled, and—though less famous than he once had been—respected."

https://lnkd.in/gsv_TXD

👉 Check out a short piece reflecting on our experimental OEWeek Ambassador Program, written by #CCCOER's Heather Blicher.
👉 We welcome feedback and thoughts on this pilot project and sustainable community participation in Open Education 🌍.

🔗 OEWeek Ambassadors: Showing Up and What We’re Learning: https://lnkd.in/eqJ5_V8d

#OpenEducation #OEWeek #OER #CommunityEngagement #HigherEducation