Skip to main content

Blog

Artificial intelligence represents a revolution comparable to agriculture and settlement, to the invention of writing or even of the printing press. These turning points transformed how we interact with the world. Today, AI touches every human activity.

To be honest, I'm still not sure what intelligence really is. In practice, I think we can say that artificial intelligence is a scientific discipline that lets machines accomplish tasks from examples rather than from instructions.

An autumn valley crossed by paths, a snow-capped mountain in the distance under a storm sky broken by light

✍️ Writings & Media

The Opinions of Talking Machines

deraison.ai · November 2026 · Warith Harchaoui

Do talking machines have political opinions? Eleven AIs take the same test: they all lean the same way. Then they play Marx, Thatcher or de Gaulle. Two AIs that nearly agree with each other then portray these figures in two different ways. At the end, take the test yourself and see where you land on the map.

✒️ 🧭 Read the article

Useful, Usable, Used

deraison.ai · October 2026 · Warith Harchaoui

Gide's sentence, "art is born of constraint, lives on struggle, dies of freedom," carried into enterprise AI. The three Us are the sieve that separates a system that lasts from a demo that shines once. Agents only sharpen it.

✒️ 🎯 Read the article

The Model Does Not Write the Number

deraison.ai · September 2026 · Warith Harchaoui

Four plain-language questions put to a company database: a number, a prediction, a forecast, a chart. The language model writes the query, the database answers and the error becomes visible again. txt2sql, txt2sql2ml with TabICL, txt2sql2viz with sprezzature.ai.

✒️ ⚖️ Read the article

While Not Good Yet

deraison.ai · September 2026 · Warith Harchaoui

The Ralph loop extends the while loop to questions no machine can settle: a readable chart, a fluent text, an honest forecast. An independent judge, a budget of tries and a written reason at every turn: that is what makes it hold.

✒️ 👁️ Read the article

An Elbow or Nothing

deraison.ai · September 2026 · Warith Harchaoui

elbow-helper detects the elbows of diminishing-returns curves, with quantified uncertainty. When the evidence is missing, it abstains. Bootstrap, BIC, null-model test: the full mathematics.

✒️ 📐 Read the article

From a Table to a Map

deraison.ai · September 2026 · Warith Harchaoui

Standpoint reads a comparison table (options as rows, criteria as columns) and draws a positioning map, axes named in the table's own language. One command does it, everything runs on your machine.

✒️ 🗺️ Read the article

The Last Slide

deraison.ai · September 2026 · Warith Harchaoui

A QR code on the last slide, the contact card lands in the audience's phones and a page counts the scans. Why and how I built qr4prez.

✒️ 📱 Read the article

Colors

deraison.ai · Warith Harchaoui

Three palettes with the same names, each swatch copied in one click: Apple's system colors for business, the Okabe-Ito palette for science, readable by people with color blindness, and a palette for print. Three tables link a color to an emotion, an idea and what psychology says about it, and everything downloads as CSV.

✒️ 🎨 Palette ↗

Dear Professor

Stanislas, 2021 · Warith Harchaoui

On March 2, 2021, my friend Quentin and I visited your preparatory class at Stanislas in Paris to thank you, on behalf of many alumni, for teaching us the “satisfaction mark”. […]

✒️ ❤️ Inspirational Teacher

Value in Data

Think Tank 4ᵉ Révolution, 2021 · Warith Harchaoui and Laurent Pantanacce

Each industrial revolution is driven by a key force (raw material, energy source or creative technology) that redefines the economy. This fourth revolution is unique because its “commodity” is abstract: data. […]

✒️ 🧐 White Paper ↗

AI for Business

Rennes School of Business, 2021-2025 · Warith Harchaoui

Bridging the gap between AI technologists and team leaders: this course helps business leaders understand AI's strategic aspects and how to implement it, reconciling science and business for corporate success.

✒️ 👨‍🏫 Executive MBA ↗

Favorite AI Books

Annotated library · Warith Harchaoui

In a fast-evolving field flooded with publications, a curated list of books I find remarkable, with brief comments, for readers eager to embark on the exciting AI journey.

✒️ 📚 Books

🗞️ Peer-Reviewed Publications

Publications reviewed by scientific peers.

Optimal-Transport-Based Machine Learning to Match Specific Patterns: Application to Detection of Regulatory Motifs in Omics Data

Journal of the Royal Statistical Society (RSS), 2024 · Thi Thanh Yen Nguyen, Warith Harchaoui, Lucile Mégret, Cloe Mendoza, Olivier Bouaziz, Christian Neri and Antoine Chambaz

We developed methods to geometrically link two point sets when those points should correspond. Our case study focuses on mRNA in genetics to understand Huntington's disease.

🗞️ 🧬 Article ↗

Generalized Mutual Information for Discriminative Clustering

Neural Information Processing Systems (NeurIPS), 2023 · Louis Ohl, Pierre-Alexandre Mattei, Charles Bouveyron, Warith Harchaoui, Mickaël Leclercq, Arnaud Droit and Frederic Precioso

Clustering groups similar objects together: customers, genes, fruits or animals. GEMINI leverages artificial neural networks in novel ways for large-scale clustering tasks. NeurIPS is the top (or second-top) conference in AI according to Google Scholar.

🗞️ ♊️ Article ↗

Learning Representations Using Neural Networks and Optimal Transport 🇫🇷🇬🇧

Ph.D. in Applied Mathematics, 2020 · Warith Harchaoui (supervised by Prof. Charles Bouveyron and Dr. Stéphane Raux)

From 2016 to 2020, this doctoral work focused on three key areas: large-scale clustering, feature importance without manual labels and confidence estimation for automated decisions. Its originality lies in combining optimal transport, neural networks and emerging generative AI.

🗞️ 🎓 Ph.D. Defense ↗

Past Talks

A few recordings