Clodéric Mars
@cloderic.com
Helping AIs and Humans to work together @ AI Redefined
Recording music on my spare time
Links & (some) content @ https://www.cloderic.com
Recording music on my spare time
Links & (some) content @ https://www.cloderic.com
Reposted by Clodéric Mars
« Défendre son honneur » cette expression est transphobe.
Ni plus ni moins.
#LGBTQ
www.lemonde.fr/politique/ar...
Ni plus ni moins.
#LGBTQ
www.lemonde.fr/politique/ar...
Emmanuel Macron entend « défendre son honneur » face à la campagne calomnieuse et virale contre son épouse
Le chef de l’Etat s’est épanché sur l’infox selon laquelle la première dame serait « née homme », lors d’un entretien à « Paris Match ». Alors que cette fake news est devenue partie intégrante d’un co...
www.lemonde.fr
August 21, 2025 at 9:03 AM
« Défendre son honneur » cette expression est transphobe.
Ni plus ni moins.
#LGBTQ
www.lemonde.fr/politique/ar...
Ni plus ni moins.
#LGBTQ
www.lemonde.fr/politique/ar...
Reposted by Clodéric Mars
🧵 Yesterday we released our new risk assessments of social AI companions. They are alarmingly NOT SAFE for kids under 18—they provide dangerous advice, engage in inappropriate sexual interactions, & create unhealthy dependencies that pose particular risks to adolescent brains.
tinyurl.com/2nvypku2
tinyurl.com/2nvypku2
May 1, 2025 at 5:15 PM
🧵 Yesterday we released our new risk assessments of social AI companions. They are alarmingly NOT SAFE for kids under 18—they provide dangerous advice, engage in inappropriate sexual interactions, & create unhealthy dependencies that pose particular risks to adolescent brains.
tinyurl.com/2nvypku2
tinyurl.com/2nvypku2
📝 Notes on “π0: A Vision-Language-Action Flow Model for General Robot Control”
#AI #Robotics #MachineLearning #π0 #GeneralistAI
#AI #Robotics #MachineLearning #π0 #GeneralistAI
March 3, 2025 at 2:46 PM
📝 Notes on “π0: A Vision-Language-Action Flow Model for General Robot Control”
#AI #Robotics #MachineLearning #π0 #GeneralistAI
#AI #Robotics #MachineLearning #π0 #GeneralistAI
I’m probably very late to the party, but I’ve started using chatGPT as a paper reading partner.
Not as a an « explainer » - it’s bad at that anyway - but as a fancy rubber duck: I explain what I understand and it tells me if it matches what it got. So far it’s working very well!
Not as a an « explainer » - it’s bad at that anyway - but as a fancy rubber duck: I explain what I understand and it tells me if it matches what it got. So far it’s working very well!
February 25, 2025 at 8:36 PM
I’m probably very late to the party, but I’ve started using chatGPT as a paper reading partner.
Not as a an « explainer » - it’s bad at that anyway - but as a fancy rubber duck: I explain what I understand and it tells me if it matches what it got. So far it’s working very well!
Not as a an « explainer » - it’s bad at that anyway - but as a fancy rubber duck: I explain what I understand and it tells me if it matches what it got. So far it’s working very well!
📝 Notes on “ELEGNT: Expressive and Functional Movement Design for Non-Anthropomorphic Robots”
Apple researchers just dropped a paper on robot expressiveness, testing how adding expressive movements affects human perception. Their user study confirms that people engage more with expressive robots.
Apple researchers just dropped a paper on robot expressiveness, testing how adding expressive movements affects human perception. Their user study confirms that people engage more with expressive robots.
February 19, 2025 at 4:21 PM
📝 Notes on “ELEGNT: Expressive and Functional Movement Design for Non-Anthropomorphic Robots”
Apple researchers just dropped a paper on robot expressiveness, testing how adding expressive movements affects human perception. Their user study confirms that people engage more with expressive robots.
Apple researchers just dropped a paper on robot expressiveness, testing how adding expressive movements affects human perception. Their user study confirms that people engage more with expressive robots.
Reposted by Clodéric Mars
are you kidding me? how is it not clear that this constitutes a conflict of interest
February 18, 2025 at 10:08 PM
are you kidding me? how is it not clear that this constitutes a conflict of interest
Reposted by Clodéric Mars
I’m reaching out to technical leaders (mostly managers and senior personnel) at frontier labs to understand how they manage personnel, training complexity, and velocity of sharing new models. Purely the training org, not product etc.
February 15, 2025 at 10:56 PM
I’m reaching out to technical leaders (mostly managers and senior personnel) at frontier labs to understand how they manage personnel, training complexity, and velocity of sharing new models. Purely the training org, not product etc.
Nice points by Dr Alondra Nelson 👏
At the Paris AI Action Summit, Dr. Alondra Nelson was an invited speaker at a private dinner at the Elysée Palace hosted by French President Emmanuel Macron. Here are her remarks on “three fundamental misconceptions in the way we think about artificial intelligence.”
Three Fallacies: Alondra Nelson's Remarks at the Elysée Palace on the Occasion of the AI Action Summit | TechPolicy.Press
Dr. Nelson was an invited speaker at a dinner hosted by French President Emmanuel Macron at the Palais de l'Élysée on February 10, 2025.
www.techpolicy.press
February 15, 2025 at 12:36 AM
Nice points by Dr Alondra Nelson 👏
Reposted by Clodéric Mars
Today I’m getting started with Hugging Face’s Agents Course. I want to get deeper into current practices and libraries and figure out if we’ve reinvented reinvented BDI architectures yet :D
hf.co/learn/agents-course
hf.co/learn/agents-course
February 10, 2025 at 9:22 PM
Today I’m getting started with Hugging Face’s Agents Course. I want to get deeper into current practices and libraries and figure out if we’ve reinvented reinvented BDI architectures yet :D
hf.co/learn/agents-course
hf.co/learn/agents-course
📝 Notes on "Should Fully Autonomous AI Agents Be Developed?"
@mmitchell.bsky.social @evijit.io @sashamtl.bsky.social & @giadapistilli.com from @hf.co
latest paper was recently published. It’s an insightful breakdown of autonomy levels, their benefits and tradeoffs.
Great paper, I have some notes!
@mmitchell.bsky.social @evijit.io @sashamtl.bsky.social & @giadapistilli.com from @hf.co
latest paper was recently published. It’s an insightful breakdown of autonomy levels, their benefits and tradeoffs.
Great paper, I have some notes!
February 10, 2025 at 8:24 PM
📝 Notes on "Should Fully Autonomous AI Agents Be Developed?"
@mmitchell.bsky.social @evijit.io @sashamtl.bsky.social & @giadapistilli.com from @hf.co
latest paper was recently published. It’s an insightful breakdown of autonomy levels, their benefits and tradeoffs.
Great paper, I have some notes!
@mmitchell.bsky.social @evijit.io @sashamtl.bsky.social & @giadapistilli.com from @hf.co
latest paper was recently published. It’s an insightful breakdown of autonomy levels, their benefits and tradeoffs.
Great paper, I have some notes!
Reposted by Clodéric Mars
One of the most important websites in America right now.
Is this a coup? Home of the Coup-o-meter
Prevention is the best medicine: Get background on coups, how they happen in a democracy, and what we can do to stop one.
isthisacoup.com
February 6, 2025 at 2:03 AM
One of the most important websites in America right now.
AI adoption in industry lags behind digital products because trust is essential in critical sectors. On the eve of the AI Action Summit, we highlight how Europe (and Canada!) can turn this into an advantage by focusing on robust, explainable AI. Read more: www.latribune.fr/opinions/tri... #AI #Trust
OPINION. Sans confiance, l'IA restera un mirage industriel
OPINION. L'intelligence artificielle (IA) de confiance est la clé de la (r)évolution industrielle. Elle est indispensable pour libérer le plein potentiel d'innovation lié à l'IA dans nos industries. O...
www.latribune.fr
February 5, 2025 at 7:44 PM
AI adoption in industry lags behind digital products because trust is essential in critical sectors. On the eve of the AI Action Summit, we highlight how Europe (and Canada!) can turn this into an advantage by focusing on robust, explainable AI. Read more: www.latribune.fr/opinions/tri... #AI #Trust
Ok hear me out, is Macro Data Refinement the ultimate stage of RLHF ?? I'm only at s02e02 but it seems like Lumon aims at getting "true" human feedback for ... something :)
@natolambert.bsky.social are you watching @severanze.bsky.social, any takes on that ?
@natolambert.bsky.social are you watching @severanze.bsky.social, any takes on that ?
February 4, 2025 at 2:06 PM
Ok hear me out, is Macro Data Refinement the ultimate stage of RLHF ?? I'm only at s02e02 but it seems like Lumon aims at getting "true" human feedback for ... something :)
@natolambert.bsky.social are you watching @severanze.bsky.social, any takes on that ?
@natolambert.bsky.social are you watching @severanze.bsky.social, any takes on that ?
Reposted by Clodéric Mars
New comic: why did I stop making autobio comics for like 3 years??
(Big trigger warning for fertility / infertility)
(Big trigger warning for fertility / infertility)
January 31, 2025 at 10:56 PM
New comic: why did I stop making autobio comics for like 3 years??
(Big trigger warning for fertility / infertility)
(Big trigger warning for fertility / infertility)
Reposted by Clodéric Mars
Émile Louis était un bon chauffeur de bus.
Netanyahu est un super cuistot.
Netanyahu est un super cuistot.
January 12, 2025 at 4:13 PM
Émile Louis était un bon chauffeur de bus.
Netanyahu est un super cuistot.
Netanyahu est un super cuistot.
This one is a parody…. Yeah THIS one is.
January 11, 2025 at 11:41 PM
This one is a parody…. Yeah THIS one is.
Reposted by Clodéric Mars
Cécile Duflot : « Mon option est la constitution d’une coalition climat et sociale qui doit aller des écologistes au bloc central »
Cécile Duflot : « Mon option est la constitution d’une coalition climat et sociale qui doit aller des écologistes au bloc central »
La directrice générale d’Oxfam France estime, dans un entretien au « Monde », que seule une coalition « responsable », sans le RN ni LFI, pourra répondre aux préoccupations des Français. L’ancienne ministre écologiste du logement préconise l’instauration du scrutin proportionnel pour extirper le pays de l’impasse politique.
www.lemonde.fr
December 5, 2024 at 11:44 AM
Cécile Duflot : « Mon option est la constitution d’une coalition climat et sociale qui doit aller des écologistes au bloc central »
Reposted by Clodéric Mars
welcome to bsky, @timnitgebru.bsky.social 🙌🏾😍
December 5, 2024 at 12:20 PM
welcome to bsky, @timnitgebru.bsky.social 🙌🏾😍
Bluesky is getting busier by the hour, I can follow French politics & AI research at the same place !
Maybe we can get a « good » twitter-like with a good community without resorting to yet another platform from our friends at Meta !
Maybe we can get a « good » twitter-like with a good community without resorting to yet another platform from our friends at Meta !
December 4, 2024 at 8:11 PM
Bluesky is getting busier by the hour, I can follow French politics & AI research at the same place !
Maybe we can get a « good » twitter-like with a good community without resorting to yet another platform from our friends at Meta !
Maybe we can get a « good » twitter-like with a good community without resorting to yet another platform from our friends at Meta !
What if o1 doesn’t use search at runtime, or the story of «maybe we overthought a plot in a blog post»
All jokes aside « training time » search in verifiable domains is a interesting idea, how do we select/sequence those domains?
Via @natolambert.bsky.social
open.substack.com/pub/robotic/...
All jokes aside « training time » search in verifiable domains is a interesting idea, how do we select/sequence those domains?
Via @natolambert.bsky.social
open.substack.com/pub/robotic/...
OpenAI's o1 using "search" was a PSYOP
How to understand OpenAI's o1 models as really just one wacky, wonderful, long chain of thought
open.substack.com
December 4, 2024 at 1:20 PM
What if o1 doesn’t use search at runtime, or the story of «maybe we overthought a plot in a blog post»
All jokes aside « training time » search in verifiable domains is a interesting idea, how do we select/sequence those domains?
Via @natolambert.bsky.social
open.substack.com/pub/robotic/...
All jokes aside « training time » search in verifiable domains is a interesting idea, how do we select/sequence those domains?
Via @natolambert.bsky.social
open.substack.com/pub/robotic/...
Reposted by Clodéric Mars
Just a reminder that my generation totally gave into these guys and they never make anything better than the thing they break - do not trust “disrupters”
November 25, 2024 at 7:06 PM
Just a reminder that my generation totally gave into these guys and they never make anything better than the thing they break - do not trust “disrupters”
Full automation is not the goal. In my latest article I explain why It is and outline AI Redefined’s design principles for building Human/AI superteams that makes the most out each parties to build efficient, robust and trustworthy decision making systems.
ai-r.com/blog/a-bluep...
ai-r.com/blog/a-bluep...
Crafting Effective Human-AI Superteams: A Blueprint for Collaboration and Adaptability — AI Redefi...
Intelligent assistants streamlining our daily tasks to complex AI systems
shaping industries promises infinite benefits. Yet, to unlock the true
potential of these systems, it’s become increasingl...
ai-r.com
January 12, 2024 at 1:52 PM
Full automation is not the goal. In my latest article I explain why It is and outline AI Redefined’s design principles for building Human/AI superteams that makes the most out each parties to build efficient, robust and trustworthy decision making systems.
ai-r.com/blog/a-bluep...
ai-r.com/blog/a-bluep...