El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile
El Mahdi El Mhamdi

@l_badikho

Ass Prof @polytechnique. Past: Senior research scientist @Google & cofounder @mamfakinch.
Book: Le Fabuleux Chantier (@EDPSciences 2019)

Secular republican.

ID: 144553035

linkhttp://elmahdielmhamdi.com calendar_today16-05-2010 16:46:25

3,3K Tweet

12,12K Followers

571 Following

El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

I don't care about birthdays, I prefer when I realise days later that mine happened without me noticing. But today, Omar Radi is spending his fourth birthday in jail. We will not forgive those who put him there, nor their allies in the so-called democratic world. Free Omar Radi

I don't care about birthdays, I prefer when I realise days later that mine happened without me noticing. But today, <a href="/OmarRADI/">Omar Radi</a> is spending his fourth birthday in jail.

We will not forgive those who put him there, nor their allies in the so-called democratic world.
<a href="/Free_Omar_Radi/">Free Omar Radi</a>
El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

Francois France Culture @ioanamanol Natacha Triou Raja Chatila il faut vous mettre d'accord "François", vous êtes un français souhaitant la privatisation du service public ou un marocain des services pas si secrets, se réjouissant de l'arrestation des défenseurs de droits humains une fois ils rentrent voir leur pays ? x.com/Francoi5211339…

El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

Francois France Culture @ioanamanol Natacha Triou Raja Chatila Najat Vallaud-Belkacem Pas besoin de nettoyer votre compte pro-régime marocain pour refaire une identité de fasciste Français, on a vu, et on sait: - monarchie marocaine, fascisme européen: même nature, grands alliés. - vous êtes fournis par les mêmes entreprises en faux comptes, dont la hongroise ;-)

El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

Yasmina Bennani La fête du trône est une fête de la monarchie, pas une fête des marocains. Darmanin, monarchiste d'action française, est le bienvenu chez les siens, monarchistes d'extrême droite, en présence de Samira Sitaïl, citoyenne française collaboratrice des barbouzes de la DGED marocaine.

El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

Yasmina Bennani Être outrés à juste titre par l'islamophobie de Darmanin/Bergé, mais pas plus par le régime de Mohammed 6, qui insère des bâtons dans les anus de ses prisonniers d'opinion (témoignage Nasser Zefzafi, actes de tortures reconnus par M6), voilà le vrai problème des franco-marocains.

Wassim (Wes) Bouaziz (@_vassim) 's Twitter Profile Photo

Want to know if a ML model was trained on your dataset? Introducing ✨Data Taggants✨! We use data poisoning to leave a harmless and stealthy signature on your dataset that radiates through trained models. Learn how to protect your dataset from unauthorized use... A 🧵

El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

Data Taggants: Excellent work on data ownership, from Wassim (Wes) Bouaziz (curent PhD @polytechnique & AI at Meta), leveraging gradient alignment/matching to sign data sets with private keys in a way that radiates through trained AI models. paper: arxiv.org/abs/2410.09101 Details in thread👇🏾

Data Taggants: Excellent work on data ownership, from <a href="/_Vassim/">Wassim (Wes) Bouaziz</a> (curent PhD @polytechnique &amp; <a href="/AIatMeta/">AI at Meta</a>), leveraging gradient alignment/matching to sign data sets with private keys in a way that radiates through trained AI models.
paper: arxiv.org/abs/2410.09101

Details in thread👇🏾
El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

When a measure becomes a target it ceases to be a good measure This adage from Marilyn Strathern known as Goodhart's law, has implications beyond economics, public policy & AI With Lê Nguyên Hoang bsky.app/profile/science4all.org we tried few years ago to formalise it, here it is finally arxiv.org/abs/2410.09638

When a measure becomes a target it ceases to be a good measure

This adage from Marilyn Strathern known as Goodhart's law, has implications beyond economics, public policy &amp; AI
With <a href="/le_science4all/">Lê Nguyên Hoang bsky.app/profile/science4all.org</a> we tried few years ago to formalise it, here it is finally arxiv.org/abs/2410.09638
El Mahdi El Mhamdi (@l_badikho) 's Twitter Profile Photo

Months before it started, it was clear that the #AIActionSummit was going to be a united Arab emirates' regime influence operation on French assets (started before 2021). Happy to have boycotted every bit of it. Sad for France's sovereignty… but will not weep for it either.

Wassim (Wes) Bouaziz (@_vassim) 's Twitter Profile Photo

Want to know if a ML model was trained on your dataset with 1 API call? See you in conferences 🙌 Excited to share that our paper Data Taggants for image data was accepted at ICLR 2025 🎉 Our follow-up on audio data, was accepted at ICASSP 2025! 🎉 Check out the details below 👇

Want to know if a ML model was trained on your dataset with 1 API call? See you in conferences 🙌

Excited to share that our paper Data Taggants for image data was accepted at ICLR 2025 🎉
Our follow-up on audio data, was accepted at ICASSP 2025! 🎉
Check out the details below 👇
Statistics Papers (@statspapers) 's Twitter Profile Photo

The Strong, Weak and Benign Goodhart's law. An independence-free and paradigm-agnostic formalisation. arxiv.org/abs/2505.23445

arXiv math.ST Statistics Theory (@mathstb) 's Twitter Profile Photo

Gilles Bareilles, et al.: On Monotonicity in AI Alignment arxiv.org/abs/2506.08998 arxiv.org/pdf/2506.08998 arxiv.org/html/2506.08998

Wassim (Wes) Bouaziz (@_vassim) 's Twitter Profile Photo

🚨New AI Security paper alert: Winter Soldier 🥶🚨 In our last paper, we show: -how to backdoor a LM _without_ training it on the backdoor behavior -use that to detect if a black-box LM has been trained on your protected data Yes, Indirect data poisoning is real and powerful!

🚨New AI Security paper alert: Winter Soldier 🥶🚨
In our last paper, we show:
-how to backdoor a LM _without_ training it on the backdoor behavior
-use that to detect if a black-box LM has been trained on your protected data

Yes, Indirect data poisoning is real and powerful!