I tweet about privacy and security, and sometimes other things...! "talking is easy, doing difficult"
Inria, France
Joined January 2013
- Tweets4.2K
- Following473
- Followers1.3K
- Likes5.3K
Castelluccia Claude retweeted
☕️ Viginum et Inria créent un prix de la recherche contre les manipulations de l’information dlvr.it/TPqlwL
Castelluccia Claude retweeted
⚠️ Contrairement à ce qui est indiqué, cette technologie n’a reçu aucun agrément de la CNIL et se doit de respecter d’elle-même le RGPD.
Retrouvez la position de la CNIL sur les caméras dites « intelligentes » ou « augmentées » dans les espaces publics 👉 cnil.fr/sites/cnil/files/ato…
🔴👁️ DOCUMENT #JT20h : dans plusieurs enseignes de supermarchés, vous êtes scannés par des caméras et l’IA note votre âge, votre style et votre sexe. Comme par hasard, vous recevez des publicités ciblées sur les écrans du magasin.
Readers added context they thought people might want to know
⚠️ Contrairement à ce qui est indiqué, cette technologie n’a reçu aucun agrément de la CNIL et se doit de respecter d’elle-même le RGPD.
Source : x.com/CNIL/status/19…
Castelluccia Claude retweeted
In 1992 Peter Ratcliffe received this rejection letter from Nature.
His findings were not "a sufficient advance in our understanding".
27 years later he won the Nobel Prize for the same discovery.
Don't lose faith in the things you believe in.
Castelluccia Claude retweeted
🤝Pour répondre aux impératifs de régulation, sécurité, transparence et d’éthique de l'#IA, la @CNIL & @Inria renforcent leur partenariat pour la protection des données, le respect de la vie privée, l’évaluation des algorithmes & des systèmes d’IA.
➕inria.fr/fr/partenariat-cnil…
Castelluccia Claude retweeted
La CNIL publie ses dernières fiches IA, en précisant les conditions d’applicabilité du RGPD aux modèles, les impératifs de sécurité et les conditions d’annotation des données d’entraînement. Elle dévoile également ses futurs travaux ici👉 cnil.fr/fr/ia-finalisation-r…
ALT Intelligence artificielle : la CNIL finalise ses recommandations sur le développement des systèmes d’IA et annonce ses futurs travaux
Castelluccia Claude retweeted
@CNIL-@Inria #Privacy Award has extended its deadline to February 21, 2025! Hurry up and submit your best #Privacy #Protection paper!
🏆 The 9th edition of the CNIL-@Inria #Privacy Award is open for submissions!
📅 Applications must be submitted before February 13th, 2025 midnight
ℹ️ More information and rules of contest on the CNIL website 👉 cnil.fr/en/launch-9th-editio…
ALT 2025 CNIL-Inria Privacy Award
Castelluccia Claude retweeted
Starting salaries at Cambridge are less than $70k for most fields. Stanford? Double that.
As someone trying to hire the best global talent to work on UK and EU projects, I can't compete with the American universities. One election won't change that.
Castelluccia Claude retweeted
[1/2] Sécurité des grands LLM : une cartographie des outils proposée par les industriels, leurs avantages et limites. #Cybermois
linc.cnil.fr/incursion-dans-…
Castelluccia Claude retweeted
Membership inference attacks against LLMs are an extremely important task for privacy/confidentiality 🕵️, copyright 📚 but also for model evaluation 💯.
Yet, despite lots of work being published on the topic 🔥, we're concerned we're not actually making progress.
Castelluccia Claude retweeted
Hundreds of generative AI models released, and only 5 really open-source?
Keeping in mind they are cultural transmission technologies, this is very very low!
What is the list of full open-source LLMs/VLMs
(= ALL datasets + training code + inference code + evaluation code are available in addition to weights)?
Examples are Bloom, Pythia, Olmo: which other models would you suggest? 🤔 cc @Thom_Wolf @BlancheMinerva @srush_nlp @HannaHajishirzi
I am trying to build this list for a course, but I guess this information may be more widely useful.
(⚠️I am NOT searching for models which are only open-weights like Llama or Mistral models)
Castelluccia Claude retweeted
Excited by the incredible support for our proposal on science- and evidence-based AI policy: understanding-ai-safety.org, with hundreds from ~150 institutions already volunteered to contribute! Let's build a safe, responsible AI future together, in CA & in our society! If you agree that:
AI policy should be science- and evidence-based; and we should prioritize advancing scientific understanding of AI risks and how to successfully mitigate them,
Please like/retweet to show support!
Castelluccia Claude retweeted
💡💡 We’ve assembled a stellar lineup of speakers to discuss AI, competition and innovation, with no less than @vestager @DAcemogluMIT & @Gorintic.
Join us on 30 September!
📅 Save the Date #AtEchelle!
Join us at 3pm on 30 September for our #GenAI webinar 🤖 “The challenges for innovation, the economy and competition”
🔴 Live on YouTube! Sign up here autoritedelaconcurrence.fr/e… #adlc
ALT generative ai: the challenges for innovation, the economy and competition
Le grand moissonnage des données personnelles - lemonde.fr/blog/binaire/2024… @Blog_Binaire
Castelluccia Claude retweeted
🚨 This is NOT an AI hallucination! 🤖
@AI_Regulation is accelerating work on generative AI, facial recognition, 🇪🇺 AI Act implementation, AI governance, & more! 🚀 We’re hiring two researchers to join our amazing team!
🌟 Don’t miss out—
👇 Check out the details and join us!
🚀 **We're Hiring! Legal Researchers in Data Protection & AI** 🧑⚖️🤖
Join @AI_Regulation @UGrenobleAlpes
- Cutting-edge AI & GDPR research 📚
- Multidisciplinary team
- Global impact 🌎
- French Alps location 🏔️
- Work-life balance: ⛷️ 🥾 🧗♀️🚵♀️
👉 ai-regulation.com/we-are-hir…
Castelluccia Claude retweeted
An Assistant Professor position in the S3 group -(@s3eurecom) at @EURECOM in vulnerability detection, analysis, exploitation, and/or remediation is still open. It is still time to send your applications!
eurecom.fr/en/job/vulnerabil…
Feel free to ask me any questions !
Castelluccia Claude retweeted
#PrivacyResearchDay Comment le droit à l'oubli pourrait il s'appliquer aux #IA génératives ? Quelles sont les stratégies pour améliorer l'auditabilité des IA ? Droit, économie et informatique seront abordés dans ce panel animé par Claude Castelluccia 👉 cnil.fr/fr/journee-de-recher…
ALT Privacy Research Day 3e édition Paris, 4 juin 2024
Castelluccia Claude retweeted
#RA2023 La CNIL a poursuivi la consolidation de liens étroits avec le monde de la recherche, en organisant notamment la 2e édition du #PrivacyResearchDay, une conférence internationale faisant dialoguer régulateurs et chercheurs.
Le rapport annuel 👉 cnil.fr/fr/la-cnil-publie-so…
Castelluccia Claude retweeted
IA : la CNIL publie ses premières recommandations sur le développement des systèmes d’intelligence artificielle cnil.fr/fr/ia-la-cnil-publie…
Castelluccia Claude retweeted
1. Today, 24 competition and consumer protection enforcement agencies issued a statement that makes clear that integrating technologists and technical experts into the policy and enforcement process is of global importance: bit.ly/3x5NlES
ALT Australian Competition and Consumer Commission, Australia Federal Competition Authority, Austria Competition Bureau, Canada Conselho Administrativo de Defesa Econômica, Brazil Directorate-General for Competition, European Commission Autorité de la concurrence, France Bundeskartellamt, Germany Autorità Garante della Concorrenza e del Mercato, Italy Japan Fair Trade Commission, Japan Competition Authority of Kenya, Kenya Malaysia Competition Commission, Malaysia Comisión Federal de Competencia Económica, Mexico The Netherlands Authority for Consumers and Markets, Netherlands Norwegian Competition Authority, Norway Autoridade da Concorrência, Portugal Competition and Consumer Commission of Singapore, Singapore Competition Commission of South Africa, South Africa Comisión Nacional de los Mercados y la Competencia, Spain Swedish Competition Authority, Sweden Competition Commission, Switzerland Turkish Competition Authority, Turkey UK Competition and Markets Authority U.S. Depart
Castelluccia Claude retweeted
How do you know if a machine learning model was trained privately?
Confidential-DPproof, developed by a team of Brave researchers and academic collaborators (led by our privacy researcher @AliShahinShams1), allows institutions to directly prove to you through the execution of a Zero Knowledge Proof protocol that they train machine learning models in a privacy-preserving manner.
Confidential-DPproof will be presented as a spotlight paper at the 12th International Conference on Learning Representations between May 7th and 11th in Vienna, Austria. #ICLR2024
Read more about Confidential-DPproof: brave.com/confidential-dppro…
You can see the full paper here: openreview.net/pdf?id=PQY2v6…
