News

September 29, 2026

A magic quadrant is an opinion, not a measurement

In 2009, a software vendor took Gartner to court over its placement in their famous quadrant. Gartner won: it is a chart but not a measurement, it is an opinion!

The Gartner Magic Quadrant is a figure with two axes. Horizontally, “completeness of vision.” Vertically, “ability to execute.” Their crossing cuts four quadrants and the logos of a market are laid on top. Upper right, the “Leaders”... and everyone looks there and wants to be there.

What a chart says is always made of two things:
• what the data timidly says
• what the author claims and puts forward as an opinion

For fun, I built a tool that spells all of that out for you. Enter your table of options scored against your criteria, click and it states your opinion in your own personal magic quadrant.

Please remember that an opinion that knows it is an opinion is more honest than an opinion that mistakes itself for a measurement.

Try it without asking permission: deraison.ai/standpoint
Code it, share it and drop me a star if I deserve one: github.com/warith-harchaoui/standpoint

September 22, 2026

The elbow of a curve is the place where it bends

Let us detect it.

How many clusters? When to stop? At which operating point do we get the most efficient power?

It is where the curve bends!

This problem has had practical solutions since at least the 1940s and I am adding my own small stone: how many bends and what happens under noise?

Try it: deraison.ai/elbow-helper
Code it and share it: github.com/warith-harchaoui/elbow-helper

A GitHub star would make my day!

September 10, 2026

A tribute to Hans Rosling: the history of every country in the world since 1950

The history of every country in the world from 1950 to today, in a handful of bubbles that light up in our heads.

In 2006, Hans Rosling walked on stage to tell us that story. The following year, Google bought his software Trendalyzer.

What he showed is not just a chart but countries moving through time:
⭕️ size = population,
🔴 color = region,
animated over decades of human development scrolling before your eyes along two axes of great explanatory power:
💰 GDP per capita
👶 share of children who live past the age of five

One of Rosling's great lessons remains painfully current: our intuition alone is a poor instrument for measuring the world.

That is the discipline I apply in companies: designing specialized teammates able to work the house data so that teams can understand, decide and act. To make that data immediately readable, those teammates can lean on sprezzature.ai, an open visualization tool I built.

Proud of our tools, we allowed ourselves a tribute to Hans Rosling by revisiting his wonderful work: sprezzature.ai/hans-rosling.html

Rosling was not remarkable because he could generate charts. He was exceptional because he could turn his data into fine understanding inside a solid frame. At a time when an AI can generate any chart in seconds, it is precisely that second part that carries the value when you have to decide.

August 26, 2026

“Valet & Artificial Intelligence,” a conference-show with Elsa Secco

📅 September 15, 2026 · Narratiiv School, Paris 11th

I’ll be on stage with Elsa Secco, a video artist, for our first-ever conference-performance: “Valet & Artificial Intelligence” as part of Le Wagon.

At the intersection of literature, art and science, we’ll explore our relationship with artificial intelligence.

We’ll discuss everything from The Phenomenology of Spirit to Jurassic Park, touching on myths and Buddhist texts along the way.

💙 The theme: What power do we choose to grant to AI?

🐘 Our goal: to celebrate science and the arts through poetry and philosophy.

📅 September 15
📍 Narratiiv School, Paris 11th arrondissement
🎟️ Registration is open; link in the comments.

Looking forward to sharing this moment with you.

Thank you to Jasky Sor for your trust.

May 6, 2026

Who's Human IA 2026: proud to be featured

I am proud to announce that in September 2026 I will be featured in WHO’S HUMAN IA, which spotlights key players in France’s artificial intelligence ecosystem.

This recognition means even more to me because I am convinced of one thing: to understand AI, we must bring together diverse perspectives.

The perspectives of artists, scientists, engineers, educators, filmmakers and entrepreneurs. At the core of this is the commitment to actively champion a French vision of artificial intelligence: rigorous, creative, ethical and human.

As early as the late 2000s, I was practicing data science using open-source tools, notably scikit-learn and OpenCV. Since then, through my experiences and the people I’ve met, I’ve seen AI transform far more than just businesses: it’s changing the ways we work, create, think, make decisions and learn.

Machine Learning, GenAI, Agentic AI and more recently Symbiotic AI: terminology evolves rapidly. How can we make sense of the interplay between theory, practice, business value and ethical responsibility without hypocrisy or posturing?

Thank you to Florence de Rochefort for bringing this to light. Thanks also to Elsa SECCO, who consistently pushes me to think outside the box.

January 11, 2026

MAYA = Most Advanced Yet Acceptable

Lipstick and Locomotives

Glamour and Innovation in AI 💄🚂

Our client has identified a problem. I can solve it with our AI. The relevance is clear.

The next question arises: How far can we go without losing support?

Raymond Loewy has continued to answer this question for me since 1951 in his autobiography and design manifesto *Never Leave Well Enough Alone*:
MAYA = Most Advanced Yet Acceptable

👉 Go as far as possible without crossing the threshold where people lose interest

Innovation isn’t a race toward magic: it’s a slider that we adjust with full awareness, without rushing through the steps

MAYA Scale for AI in business with increasing levels of risk

1. Explain: detects / classifies / summarizes / contextualizes. Humans decide

2. Suggest: proposes options + trade-offs. Humans have the final say

3. Draft: prepares a deliverable (email, code, analysis, plan). Saves time

4. Approve: executes after validation. Active traceability

5. Autonomy: acts independently, but within defined boundaries (rules, budget, audit)

💡 Successful AI projects go ONE step beyond the comfort zone

🎯 Golden rule: the higher the cost of error, the lower the level of autonomy must be.

To increase it, we don’t add “magic”: we add control

January 10, 2026

Ecology. Sovereignty. Bias. In business… all hot air?

As long as it remains mere lip service, yes

As Head of AI, I talk about costs, the company’s long-term viability and the law and that’s how I uphold these values

🌱 “Eco-responsible AI”

❌ It’s nice, but ineffective
Fewer calculations = less money spent = less CO₂

Let’s take advantage of it!
🔗 lnkd.in/eX3YbE5x

For the sake of the Earth, whenever possible:
Bill first, environmental impact second

🏳️ “Sovereign AI”
❌ It’s patriotic but ineffective
Let’s talk about the company’s long-term viability
Even the French government too often chooses Microsoft and Google

In 2000, Toys “R” Us signed a ten-year exclusivity agreement with Amazon, outsourcing its online sales and delaying the launch of its own e-commerce platform.
When Amazon expanded its network of suppliers, Toys “R” Us won the lawsuit but had already lost its momentum toward e-commerce
🔗 lnkd.in/ecn5a_98

Let’s be pragmatic:
Corporate sovereignty > sovereignty in general

⚖️ “Ethical and inclusive AI”
❌ It’s noble but ineffective
Concrete examples:
• Google (2015) classified a photo of an African American couple as “gorillas” in a public demo
• Meta (2021), same thing in a video (“primates”)
Biased training data (overrepresentation of adult white men) → statistical disaster
🔗 lnkd.in/e-4EEduE


What's Changing the Game: Reputational and Legal Risk

March 31, 2025

November 2022: one leg less, a passion untouched

November 2022. The doctors are unanimous: “We have to amputate this right leg” that’s causing me so much trouble.

A few weeks earlier, Elsa SECCO 👾 spoke to me with rare conviction. Her encouragement helped me grow and make a decision that, nevertheless, revolted me. Together, we turned down startup funding we’d already secured because of this leg that had become unusable, as the doctors explained to me.

My memories of that day are fading. The decision was made. I went to the hospital alone, without music in my ears, even though I usually listen to Satie, Little Richard or whatever my ears can pick up.

Not out of mental strength. Out of overwhelm. Too much tension.

🔇 I needed silence.

🧠 My mind, meanwhile, drifts toward what I love to do: artificial intelligence research. ChatGPT-3.5 has just been released. On my Mac sitting on my hospital bed, it feels like an earthquake.
Nicolas Renard, my boss at Jellysmack, is filling in for me on short notice in Rennes and Andrea Colonna agrees and tells me he’d love to come too! Nicolas is making the round trip from Toulouse to Paris to Rennes because he thinks it’s important. Marc Caillet checks in with me from time to time with updates from work and a few jokes.

These guys are so classy!

📐 I keep reading papers on arXiv. In my suitcase, there’s a serious book: the draft that Francis Bach suggested I read, his “Learning Theory from First Principles” (two years before its release). This book is a real page-turner! I pick up my pen to redo a calculation. Feeling the intuition and the joy of understanding.

📞 I call Mohamed CHELALI: “How’s it going with speaker separation?” He tells me it’s working, a little surprised that I’m thinking about this just for fun 🤓

A few months later, after working with a medical team that’s nothing short of magical, I’m walking and running on a straight metal leg!

I can travel without the daily care of a nurse worried about a potential risk of infection. So today, when I step out of a business meeting, in a suit, a crisp white shirt… 👉 I’m showing off.
Earlier today, while the weather was nice in Paris, I asked a passerby to take a photo of me with their iPhone, just to make my friends laugh.

People sometimes ask me:
“Why did you name your AI company deraison.ai?”
Because I love this soundscape: folly, triumphant reason, derision.

I believe we should never take ourselves seriously, but we must take everything we have the privilege of creating abnormally seriously.

March 20, 2025

scikit-learn was born of a constraint and my profession with it

✊ I maintain that my job as a Data Scientist was born in a brain imaging laboratory 🧠 in France. A physicist wanted a free tool that would beat a widespread and very expensive piece of software 💰

scikit-learn, created in 2007 🎉, is today one of the pillars of Open Source Data Science thanks to Gaël Varoquaux. Many others have evidently followed him and many more keep arriving all over the world 🌍

🔹 Gaël used Matlab 📊, then the reference in applied science. But when he set out to automate complex experiments, zero gravity flights among them 🚀, he found Matlab unreliable ☹️

🔹 While he worked on brain imaging at NeuroSpin 🧠, some labs were paying more for Matlab licenses than for their hardware 💸

🔹 Gaël decided to do better and for free 🤯, through one creative philosophy: Open Source AI

🏢 scikit-learn succeeded so well that it sits at the heart of enterprise AI: feeding data to algorithms to create value

André Gide wrote: “Art is born of constraint, lives on struggle and dies of freedom”

In everything I undertake, I try to name my constraints, my struggles and my freedoms

March 17, 2025

I work in AI and yet tonight I'm at the Senate

Here’s why 👇

🏛️ AI isn’t just a technical issue: it’s a political, economic and strategic one

So I’m here to listen to my friend Alkéos Michaïl, a PhD from the 4ème Révolution think tank and my boss Yann Lechelle at :probabl.

🧑‍🔬 I’ve always been passionate about mathematics and artificial intelligence, even before it became trendy… and sometimes to the point where I’d rather chat with researchers than go to class! 😇

💡 Why?
Because I want AI that’s rooted in the real economy and creates tangible value 🌍

👣 My journey in AI
🚗 Oscaro: Back when industrial data science was practically nonexistent (in 2014!), I saw how it could impact a sector as concrete as auto parts
📺 Jellysmack: Diving into a Corsican-American unicorn, I discovered the world of digital content and how AI shapes perceptions
🤖 :probabl. Today, I work alongside the creators of scikit-learn, who shaped my career by providing what many consider an essential pillar of open-source AI

🌍 At :probabl., we hold a strong conviction: AI must remain open and controllable
✅ Open Source, not a black box that’s impossible to maintain or understand ◼️
✅ Open Weights, not opaque SaaS that enslaves you to an exorbitantly expensive company 🏢
✅ Open Data, not the reckless and blind exploitation of data 🛡️
✅ Sovereign AI that liberates rather than confines you 💪

🤝 At :probabl., We help companies take (or regain) control of their AI
🔍 Model audits & trustworthy AI
⚙️ Scaling open-source solutions
🔝 Building the skills of “non-AI” professionals by putting your business lines at the center
🤔 Understanding the fundamentals to master AI challenges

💡 I’ll be sharing plenty of stories and success stories in these industries
🏦 in finance
🏥 in healthcare
🏭 in logistics and manufacturing
📰 in media
so follow me on LinkedIn

November 19, 2024

First day at probabl.!

Today, I’m joining a company whose people I’ve admired for at least 15 years: ever since that first day when I typed the words “import sklearn” 😌

Chronologically, thank you to Yann Lechelle, Olivier Grisel, Nicolas Delaforge, Franck Charras, Charlène Bizollon, Emma Tysinger, Pierre-René Lherisson, Freddy Limpens and everyone else for your trust. Finally getting to wear the famous blue sweatshirt with orange squares (the one I loved so much at the PyData conference a few weeks ago) is both an honor and a responsibility 🙏

Since leaving Jellysmack in May 2024, after nearly three exciting years, I’ve asked myself a thousand questions about my career path and my entrepreneurial, academic and scientific aspirations. These reflections were fueled by wonderful conversations with people I love (you’ll know who you are (hint: there was often a meal or a coffee involved in our stories 😋)

What am I going to do?
👉 Contribute to open-source artificial intelligence from France, with unanimous international recognition (and yes, ask the experts and they’ll confirm it 😏).
👉 Wearing many hats: from theory to practice, from need to product, from science to applications, from problem to solution… and from dream to reality with funding 💡

Why?
This opportunity brings together two aspirations:
1️⃣ Excellence alongside the best: it’s as simple as that
2️⃣ The satisfaction of solving real-world problems to serve clients

Moreover, this mission is part of a broader challenge that goes beyond me: Open Source in AI liberates knowledge, just as literacy transformed our societies in the past. Whereas some once doubted the economic value of mass education, we know just how immense an impact those choices have on both today and tomorrow.

At :probabl., I take on the challenge of imagining, listening, understanding and rearticulating (mathematically, computationally and commercially) what the client needs, with panache, I hope (for the pleasure of using such a French word 🇫🇷)

If you’ve read this far, you should send me a message or leave a comment: I’d love to chat! 🤝