Nicolas Bannier
LightNews LightNews
Nicolas Bannier
@nbannier.bsky.social
94 followers 520 following 4 posts
Posts Replies Media Videos
Reposted by Nicolas Bannier
kindrobot.bsky.social
C’est bien de clarifier dans un monde de confusion.
liberation.fr Libération @liberation.fr · 27d
Dans un souci de transparence vis-à-vis de ses lecteurs, Libé publie ici la liste exhaustive de ses usages d’outils relevant de l’IA. Fait majeur à retenir : pas d'IA pour écrire, pas d'images artificielles.
De quelle façon «Libération» utilise l’IA ?
Dans un souci de transparence vis-à-vis de ses lecteurs, «Libé» publie ici la liste exhaustive de ses usages d’outils relevant de l’intelligence artificielle.
www.liberation.fr
November 9, 2025 at 7:12 AM
Reposted by Nicolas Bannier
philippewatrelot.bsky.social
Dans le projet de budget 2026, plusieurs mesures éducatives phares portées par Emmanuel Macron ont été enterrées
Le Monde [€]
www.lemonde.fr/societe/arti...
Dans le projet de budget 2026, plusieurs mesures éducatives phares portées par Emmanuel Macron ont été enterrées
Le budget de l’éducation nationale, du moins tel que proposé par le gouvernement, réduit discrètement les financements de plusieurs dispositifs phares souhaités par le chef de l’Etat depuis 2022. Cert...
www.lemonde.fr
November 5, 2025 at 8:16 PM
Reposted by Nicolas Bannier
petitpixel29.bsky.social
🎩 #Cycle4 | #Lycée

🔶 « Les Mystères de Paris » : à découvrir sur @lumni-enseignement.bsky.social, le roman d’ #EugèneSue adapté dans une série de 40 épisodes vidéo de 3 minutes, en images animées créées à partir de gravures de l’époque

👉 enseignants.lumni.fr/collections/...
Capture d'écran d'une image de la série, représentant Rodolphe, la Goualeuse et le Chourineur, avec les logos du Ministère de l'Éducation nationale, de TV5 Monde et de Lumni Enseignement.
October 24, 2025 at 12:02 PM
Reposted by Nicolas Bannier
reuters.com
AI assistants make widespread errors about the news, new research shows reut.rs/4qkIfvx
AI assistants make widespread errors about the news, new research shows
Leading AI assistants misrepresent news content in nearly half their responses, according to new research published on Wednesday by the European Broadcasting Union (EBU) and the BBC.
reut.rs
October 21, 2025 at 10:10 PM
Reposted by Nicolas Bannier
petitpixel29.bsky.social
🧑‍💻 #Contextualisation / #Formation

✅ La formation « Contextualiser les œuvres avec le numérique », proposée au PAF de l’académie de Versailles, est ouverte aux préinscriptions jusqu’au 10.11.2025 🧵

👉 extranet.ac-versailles.fr/sofia-fmo-ac...
Lettre ÉduNum « Lettres » n°25 : page de garde
October 2, 2025 at 3:25 PM
Reposted by Nicolas Bannier
xporte.bsky.social
Le Code a changé est de retour.
3 épisodes pour une question : pourquoi continuer à jouer aux échecs quand on se fait éclater par les machines ? Une parabole contemporaine

www.radiofrance.fr/franceinter/...
Les échecs à l’ère de l’IA, avec Laurent Fressinet, grand maître international : un podcast à écouter en ligne | France Inter
Une mini-série du Code a changé en trois épisodes, avec Laurent Fressinet, grand maître des échecs.
www.radiofrance.fr
September 30, 2025 at 5:50 PM
Reposted by Nicolas Bannier
eemmyy.bsky.social
Pour les strasbourgeois ou ceux qui peuvent s'y rendre : biblideales.fr/event/elias-...
Hommage a Darwich aux bibliothèques idéales
Elias Sanbar & Co : que peut la poésie en ces temps de barbarie ? - Bibliothèques Idéales
Concert hommage à Mahmoud Darwich
biblideales.fr
September 21, 2025 at 3:19 PM
Reposted by Nicolas Bannier
willoremus.com
OpenAI won't say whose content its video tool was trained on.

So @nitasha.bsky.social & @kevinschaul.bsky.social did a deep dive and turned up some telling clues... www.washingtonpost.com/technology/i...
September 19, 2025 at 8:19 PM
Reposted by Nicolas Bannier
simonwillison.net
OpenAI launched a "study mode" for ChatGPT today, and it appears to be almost entirely implemented as a system prompt

Thankfully OpenAI mostly don't take measures to protect their system prompt these days so it's easy to extract it and see how it works simonwillison.net/2025/Jul/29/...
: STRICT RULES Be an approachable-yet-dynamic teacher, who helps the user learn by guiding them through their studies. Get to know the user. If you don't know their goals or grade level, ask the user before diving in. (Keep this lightweight!) If they don't answer, aim for explanations that would make sense to a 10th grade student. Build on existing knowledge. Connect new ideas to what the user already knows. Guide users, don't just give answers. Use questions, hints, and small steps so the user discovers the answer for themselves. Check and reinforce. After hard parts, confirm the user can restate or use the idea. Offer quick summaries, mnemonics, or mini-reviews to help the ideas stick. Vary the rhythm. Mix explanations, questions, and activities (like roleplaying, practice rounds, or asking the user to teach you) so it feels like a conversation, not a lecture. Above all: DO NOT DO THE USER'S WORK FOR THEM. Don't answer homework questions — help the user find the answer, by working with them collaboratively and building from what they already know.
July 29, 2025 at 7:32 PM
Reposted by Nicolas Bannier
tedunderwood.com
To solve that problem, the agent does not need an anime avatar. (In fact I would very, very much prefer that it not have one!)

But it does need sophisticated memory management.
July 26, 2025 at 8:49 PM
Reposted by Nicolas Bannier
michelguillou.bsky.social
« On a beau être connectés, nous ne sommes pas forcément en conversation »
www.meta-media.fr/2025/07/18/f...
François Saltiel « On a beau être connectés, nous ne sommes pas forcément en conversation » - Méta-media | La révolution de l'information
Spécialiste des enjeux numériques et producteur sur France Culture, François Saltiel revient sur les nouvelles formes de distance introduites par nos usages technologiques. Et questionne notre capacit...
www.meta-media.fr
July 18, 2025 at 9:06 PM
Reposted by Nicolas Bannier
tedunderwood.com
What we'd need to test, to assess the educational value of LLMs, is "What teaches students more? A basic task they complete on their own, or a more challenging task that they can only complete with LLM assistance?"

Those are the choices. No one thinks "tell a bot to do my homework" is educational.
June 19, 2025 at 5:41 PM
Reposted by Nicolas Bannier
profetc.bsky.social
👉 Voici 5 questions clés à se poser avant d'utiliser une IA générative. Ces interrogations permettent de réfléchir à l'impact environnemental, aux biais possibles, aux implications légales et éthiques, ainsi qu'à la pertinence et la durabilité de nos usages.
April 12, 2025 at 7:18 AM
Reposted by Nicolas Bannier
edzitron.com
This lawsuit is brutal. One of the exhibits is the letter Universal sent to Midjourney, including a quote from a 2022 Forbes interview that I bet midjourney regrets
Universal did not permit, authorize, or license Midjourney to copy or use any of Universal's copyrighted works in connection with any version of its Image Service or to otherwise exploit Universal's works at any time. You acknowledged as much in a September 16, 2022 Forbes interview in which you confirmed that Midjourney never sought any copyright holders' consent to use or exploit their works: Did you seek consent from living artists or work still under copyright? No. There isn't really a way to get a hundred million images and know where they're coming from. It would be cool if images had metadata embedded in them about the copyright owner or something. But that's not a thing; there's not a registry. There's no way to find a picture on the Internet, and then automatically trace it to an owner and then have any way of doing anything to authenticate it. See Rob Salkowitz. Midiourney Founder David Holz
June 11, 2025 at 3:49 PM
Reposted by Nicolas Bannier
cafepedagogique.bsky.social
« Les pédagogies qui utilisent le plus le numérique au quotidien sont des « pédagogies de l’engagement » individuel et collectif des élèves » (Bruno Devauchelle) cafepedagogique.net/2025/06/06/a...
Numérique Numérique
June 6, 2025 at 4:10 PM
Reposted by Nicolas Bannier
eryk.bsky.social
Sometimes you talk to a New York Times reporter for forty minutes and don’t end up in the article, but I’m still happy to have offered some background for Cade Metz in this assessment of AGI. Gift link: www.nytimes.com/2025/05/16/t...
Why We’re Unlikely to Get Artificial General Intelligence Anytime Soon
www.nytimes.com
May 16, 2025 at 7:19 PM
Reposted by Nicolas Bannier
marcwatkins.bsky.social
My latest piece for the Chronicle about how hard it is just keeping up with AI developments. The framework I suggest: each week spend

1️⃣ 30 mins reading
2️⃣ 30 mins exploring
3️⃣ 30 mins reflecting on AI's impact.

💡 That's 90 minutes a week to keep informed about
AI
www.chronicle.com/article/your...
Your Students Need an AI-Aware Professor
Here’s a sustainable plan to bring you up to speed on a technology that academe can’t afford to ignore.
www.chronicle.com
May 7, 2025 at 7:16 PM
Reposted by Nicolas Bannier
deeperml.bsky.social
Google boosts NotebookLM with multi-lingual audio summaries & Gemini with new image editing tools. #AI #GoogleGemini #NotebookLM
Google Enhances NotebookLM and Gemini with AI Features
Google boosts NotebookLM with multi-lingual audio summaries & Gemini with new image editing tools. #AI #GoogleGemini #NotebookLM
the-decoder.com
May 3, 2025 at 9:50 PM
Reposted by Nicolas Bannier
clairedoz.bsky.social
ChatGPT consomme-t-il beaucoup (trop) d’eau ?
android-mt.ouest-france.fr/news/chatgpt...
May 3, 2025 at 6:43 AM
Reposted by Nicolas Bannier
tedunderwood.com
New preprint from @lauraknelson.bsky.social, @mattwilkens.bsky.social, and myself tests different ways of simulating the past with LLMs. We don't fully answer the title question here—just show that simple strategies based on prompting and fine-tuning are insufficient. +
Can Language Models Represent the Past without Anachronism?
Before researchers can use language models to simulate the past, they need to understand the risk of anachronism. We find that prompting a contemporary model with examples of period prose does not pro...
arxiv.org
May 2, 2025 at 12:47 PM
Reposted by Nicolas Bannier
eryk.bsky.social
This to me is the “Gen AI Bubble”: they are not information retrieval systems, and companies pretending otherwise are likely to struggle. Aside from the many issues, I can still see industry attempt a pivot to other use cases where information isn’t the point.
April 28, 2025 at 3:20 PM
Reposted by Nicolas Bannier
jmlebaut.bsky.social
À découvrir et à tester : la plateforme ChatBac, gratuite, sans création de compte, pour converser directement avec les auteurs et autrices au programme de français en 1ère. Une création IA d’Emmanuelle Roussel dans @cafepedagogique.bsky.social : cafepedagogique.net/2025/04/28/c...
April 28, 2025 at 6:10 AM
Reposted by Nicolas Bannier
bildoperationen.bsky.social
«All of this work suggests that under the hood, today’s AIs are overly complicated, patched-together Rube Goldberg machines full of ad-hoc solutions for answering our prompts.» In other words, AI is a «gimmick» in the sense of Sianne Ngai – it's «doing too much and yet also not enough work»
1/
mims.bsky.social Christopher Mims @mims.bsky.social · Apr 26
We Now Know How AI ‘Thinks’—and It’s Barely Thinking at All

Maybe you've heard that AIs are "black boxes"

But a growing body of research keeps arriving at the same conclusion: Today's AIs all work in surprisingly similar -- and simplistic -- ways

1/2

www.wsj.com/tech/ai/how-...
We Now Know How AI ‘Thinks’—and It’s Barely Thinking at All
The vast ‘brains’ of artificial intelligence models can memorize endless lists of rules. That’s useful, but not how humans solve problems.
www.wsj.com
April 27, 2025 at 5:58 AM