Pierre-Yves Oudeyer
@pyoudeyer.bsky.social
Research director @Inria, Head of @flowersInria
lab, prev. @MSFTResearch @SonyCSLParis
Artificial intelligence, cognitive sciences, sciences of curiosity, language, self-organization, autotelic agents, education, AI and society
http://www.pyoudeyer.com
lab, prev. @MSFTResearch @SonyCSLParis
Artificial intelligence, cognitive sciences, sciences of curiosity, language, self-organization, autotelic agents, education, AI and society
http://www.pyoudeyer.com
Major new system and result from the team: SOAR is an open-source self-improving genAI system pushing the frontier of ARC performances using program synthesis
It relies on using LLMs as self-improving smart operators for evolutionary search
It relies on using LLMs as self-improving smart operators for evolutionary search
Introducing SOAR 🚀, a self-improving framework for prog synth that alternates between search and learning (accepted to #ICML!)
It brings LLMs from just a few percent on ARC-AGI-1 up to 52%
We’re releasing the finetuned LLMs, a dataset of 5M generated programs and the code.
🧵
It brings LLMs from just a few percent on ARC-AGI-1 up to 52%
We’re releasing the finetuned LLMs, a dataset of 5M generated programs and the code.
🧵
July 16, 2025 at 3:25 PM
Major new system and result from the team: SOAR is an open-source self-improving genAI system pushing the frontier of ARC performances using program synthesis
It relies on using LLMs as self-improving smart operators for evolutionary search
It relies on using LLMs as self-improving smart operators for evolutionary search
Using LLMs to advance the cognitive science of collectives
Very interesting new paper by @sucholutsky.bsky.social
Katherine Collins @norijacoby.bsky.social @billdthompson
@roberthawkins.bsky.social
arxiv.org/pdf/2506.00052
Very interesting new paper by @sucholutsky.bsky.social
Katherine Collins @norijacoby.bsky.social @billdthompson
@roberthawkins.bsky.social
arxiv.org/pdf/2506.00052
June 10, 2025 at 12:18 PM
Using LLMs to advance the cognitive science of collectives
Very interesting new paper by @sucholutsky.bsky.social
Katherine Collins @norijacoby.bsky.social @billdthompson
@roberthawkins.bsky.social
arxiv.org/pdf/2506.00052
Very interesting new paper by @sucholutsky.bsky.social
Katherine Collins @norijacoby.bsky.social @billdthompson
@roberthawkins.bsky.social
arxiv.org/pdf/2506.00052
Reposted by Pierre-Yves Oudeyer
Le travail de Samuel Bianchini à partir des algorithmes de @pyoudeyer.bsky.social mis en lumière par le Jeu de paume dans son expo IA
June 6, 2025 at 4:29 PM
Le travail de Samuel Bianchini à partir des algorithmes de @pyoudeyer.bsky.social mis en lumière par le Jeu de paume dans son expo IA
Humans' ability to invent their own games & goals is at the core of open-ended learning.
Understanding and modeling computationally how they do it would be enlightening to understand better human cognition and build open-ended AI
Great step in this direction in new paper by Guy Davidson et al.
Understanding and modeling computationally how they do it would be enlightening to understand better human cognition and build open-ended AI
Great step in this direction in new paper by Guy Davidson et al.
May 28, 2025 at 2:20 PM
Humans' ability to invent their own games & goals is at the core of open-ended learning.
Understanding and modeling computationally how they do it would be enlightening to understand better human cognition and build open-ended AI
Great step in this direction in new paper by Guy Davidson et al.
Understanding and modeling computationally how they do it would be enlightening to understand better human cognition and build open-ended AI
Great step in this direction in new paper by Guy Davidson et al.
Reposted by Pierre-Yves Oudeyer
I reviewed "These Strange New Minds: How AI Learned to Talk and What It Means" by Chris Summerfield.
melaniemitchell.me/EssaysConten...
melaniemitchell.me/EssaysConten...
May 19, 2025 at 6:36 PM
I reviewed "These Strange New Minds: How AI Learned to Talk and What It Means" by Chris Summerfield.
melaniemitchell.me/EssaysConten...
melaniemitchell.me/EssaysConten...
Cool
Published my WebGPU Particle Life simulation online! Obviously it requires WebGPU support, and it's quite resource-demanding, so I've set the particle count to 64k for now. Also it probably doesn't work on mobile; will fix later.
Enjoy: lisyarus.github.io/webgpu/parti...
Enjoy: lisyarus.github.io/webgpu/parti...
Particle Life 2D
lisyarus.github.io
May 15, 2025 at 11:40 AM
Cool
Reposted by Pierre-Yves Oudeyer
Reposted by Pierre-Yves Oudeyer
Here is a cool Flow-Lenia simulation, which is a continuous cellular automaton with two distinct features:
1. Mass conservation - sum of all activations is constant through time
2. Localized update rules - enables simulations with differently behaving matter on the same grid
1. Mass conservation - sum of all activations is constant through time
2. Localized update rules - enables simulations with differently behaving matter on the same grid
May 9, 2025 at 9:17 PM
Here is a cool Flow-Lenia simulation, which is a continuous cellular automaton with two distinct features:
1. Mass conservation - sum of all activations is constant through time
2. Localized update rules - enables simulations with differently behaving matter on the same grid
1. Mass conservation - sum of all activations is constant through time
2. Localized update rules - enables simulations with differently behaving matter on the same grid
Reposted by Pierre-Yves Oudeyer
Very happy that our paper received both the EvoApps best paper and the EvoStar best student paper! Congrats team!!
Emergent Kin Selection of Altruistic Feeding via Non-Episodic Neuroevolution
arxiv.org/abs/2411.10536
Emergent Kin Selection of Altruistic Feeding via Non-Episodic Neuroevolution
arxiv.org/abs/2411.10536
May 12, 2025 at 3:45 PM
Very happy that our paper received both the EvoApps best paper and the EvoStar best student paper! Congrats team!!
Emergent Kin Selection of Altruistic Feeding via Non-Episodic Neuroevolution
arxiv.org/abs/2411.10536
Emergent Kin Selection of Altruistic Feeding via Non-Episodic Neuroevolution
arxiv.org/abs/2411.10536
Reposted by Pierre-Yves Oudeyer
Check out @nicolasyax.bsky.social
thread about our paper (co-supervised by @pyoudeyer.bsky.social) where we show that evolutionary tree reconstruction can be successfully applied to map LLMs to map relations and predict their performance! Currently at @iclr-conf.bsky.social
thread about our paper (co-supervised by @pyoudeyer.bsky.social) where we show that evolutionary tree reconstruction can be successfully applied to map LLMs to map relations and predict their performance! Currently at @iclr-conf.bsky.social
April 29, 2025 at 11:38 AM
Check out @nicolasyax.bsky.social
thread about our paper (co-supervised by @pyoudeyer.bsky.social) where we show that evolutionary tree reconstruction can be successfully applied to map LLMs to map relations and predict their performance! Currently at @iclr-conf.bsky.social
thread about our paper (co-supervised by @pyoudeyer.bsky.social) where we show that evolutionary tree reconstruction can be successfully applied to map LLMs to map relations and predict their performance! Currently at @iclr-conf.bsky.social
Generative AI is a cultural transmission technology:
it plays a growing role in generation, selection and transmission of ideas/opinions in human society 🧠🔄🌐
And yet we understand very little of this dynamics at this point 🤔❓
A step forward is our #ICLR2025 paper !👇
it plays a growing role in generation, selection and transmission of ideas/opinions in human society 🧠🔄🌐
And yet we understand very little of this dynamics at this point 🤔❓
A step forward is our #ICLR2025 paper !👇
April 25, 2025 at 1:41 PM
Generative AI is a cultural transmission technology:
it plays a growing role in generation, selection and transmission of ideas/opinions in human society 🧠🔄🌐
And yet we understand very little of this dynamics at this point 🤔❓
A step forward is our #ICLR2025 paper !👇
it plays a growing role in generation, selection and transmission of ideas/opinions in human society 🧠🔄🌐
And yet we understand very little of this dynamics at this point 🤔❓
A step forward is our #ICLR2025 paper !👇
Imagine you can compute a (very) cheap behavioural signatures for LLMs and from it
1) infer their history (which previous models they derived from)
2) build large maps to navigate spaces of 100s of models
3) predict (coarsely) their performances in benchmarks
That's PhyloLM 🚀
1) infer their history (which previous models they derived from)
2) build large maps to navigate spaces of 100s of models
3) predict (coarsely) their performances in benchmarks
That's PhyloLM 🚀
🔥Our paper PhyloLM got accepted at ICLR 2025 !🔥
In this work we show how easy it can be to infer relationship between LLMs by constructing trees and to predict their performances and behavior at a very low cost with @stepalminteri.bsky.social and @pyoudeyer.bsky.social ! Here is a brief recap ⬇️
In this work we show how easy it can be to infer relationship between LLMs by constructing trees and to predict their performances and behavior at a very low cost with @stepalminteri.bsky.social and @pyoudeyer.bsky.social ! Here is a brief recap ⬇️
April 24, 2025 at 1:21 PM
Imagine you can compute a (very) cheap behavioural signatures for LLMs and from it
1) infer their history (which previous models they derived from)
2) build large maps to navigate spaces of 100s of models
3) predict (coarsely) their performances in benchmarks
That's PhyloLM 🚀
1) infer their history (which previous models they derived from)
2) build large maps to navigate spaces of 100s of models
3) predict (coarsely) their performances in benchmarks
That's PhyloLM 🚀
We are starting a new major research project investigating the joint development of curiosity and metacognition in adolescents, where the Flowers lab will collaborate with @mjgruber.bsky.social 's lab at Univ. Cardiff, and Yana Fandakova at Univ. Trier !!! Open postdocs positions, see below 👇
Interested in curiosity and metacognition in childhood and adolescence? Join us at Cardiff University (UK) for a Postdoc position in an international collaboration with @yanafandakova.bsky.social (Germany) and @pyoudeyer.bsky.social (France). Please repost! #DevPsych #PsychSciSky #neuroskyence
Job Search | Cardiff University
Search for Internet job results.
krb-sjobs.brassring.com
April 16, 2025 at 9:54 AM
We are starting a new major research project investigating the joint development of curiosity and metacognition in adolescents, where the Flowers lab will collaborate with @mjgruber.bsky.social 's lab at Univ. Cardiff, and Yana Fandakova at Univ. Trier !!! Open postdocs positions, see below 👇
Why Self-Determination Theory Needs Computational Modelling: The Case of Competence and Optimal Challenge
by @codingconduct.cc E. Lintunen N. Aly @creativeendvs.bsky.social
osf.io/n6x8s/downlo...
by @codingconduct.cc E. Lintunen N. Aly @creativeendvs.bsky.social
osf.io/n6x8s/downlo...
April 10, 2025 at 1:24 PM
Why Self-Determination Theory Needs Computational Modelling: The Case of Competence and Optimal Challenge
by @codingconduct.cc E. Lintunen N. Aly @creativeendvs.bsky.social
osf.io/n6x8s/downlo...
by @codingconduct.cc E. Lintunen N. Aly @creativeendvs.bsky.social
osf.io/n6x8s/downlo...
Forcing Diffuse Distributions out of Language Models
"When LLMs are used for real-world tasks where diversity of outputs is crucial their inability to produce
diffuse distributions over valid choices is a major hurdle"
New method to address this challenge by Zhang et al.:
arxiv.org/pdf/2404.10859
"When LLMs are used for real-world tasks where diversity of outputs is crucial their inability to produce
diffuse distributions over valid choices is a major hurdle"
New method to address this challenge by Zhang et al.:
arxiv.org/pdf/2404.10859
April 3, 2025 at 9:15 AM
Forcing Diffuse Distributions out of Language Models
"When LLMs are used for real-world tasks where diversity of outputs is crucial their inability to produce
diffuse distributions over valid choices is a major hurdle"
New method to address this challenge by Zhang et al.:
arxiv.org/pdf/2404.10859
"When LLMs are used for real-world tasks where diversity of outputs is crucial their inability to produce
diffuse distributions over valid choices is a major hurdle"
New method to address this challenge by Zhang et al.:
arxiv.org/pdf/2404.10859
Reposted by Pierre-Yves Oudeyer
Accepting “the bitter lesson” and embracing the brain’s complexity by Dyer and Richards. Great article! www.thetransmitter.org/neuroai/acce...
» Accepting “the bitter lesson” and embracing the brain’s complexity Close menuMenuClose menuSpectrum logoYouTubeBlueskyMastodonYouTubeBlueskyMastodon
www.thetransmitter.org
March 26, 2025 at 4:09 PM
Accepting “the bitter lesson” and embracing the brain’s complexity by Dyer and Richards. Great article! www.thetransmitter.org/neuroai/acce...
Reposted by Pierre-Yves Oudeyer
"The Appeal of Insight: Why Riddles and Whodunits Captivate Us" - These studies contribute to a growing literature in which the cultural success of fictional stories is explained by their ability to tap into specific cognitive mechanisms.
osf.io/preprints/ps...
osf.io/preprints/ps...
OSF
osf.io
March 26, 2025 at 6:44 AM
"The Appeal of Insight: Why Riddles and Whodunits Captivate Us" - These studies contribute to a growing literature in which the cultural success of fictional stories is explained by their ability to tap into specific cognitive mechanisms.
osf.io/preprints/ps...
osf.io/preprints/ps...
Reposted by Pierre-Yves Oudeyer
Extreme stances about AI and LLMs in particular (be them deflationary or inflationary) are very likely wrong, valid and certainly not good for an healthy debate. But where do they come from? We (with @giadapistilli.com) argue that common cognitive biases may play a role:
osf.io/preprints/ps...
osf.io/preprints/ps...
March 26, 2025 at 7:33 AM
Extreme stances about AI and LLMs in particular (be them deflationary or inflationary) are very likely wrong, valid and certainly not good for an healthy debate. But where do they come from? We (with @giadapistilli.com) argue that common cognitive biases may play a role:
osf.io/preprints/ps...
osf.io/preprints/ps...
Reposted by Pierre-Yves Oudeyer
Humans excel at estimating their own competence and progress on tasks. But can LLM agents do the same?
With 🧭MAGELLAN, our agent predicts its own learning progress across vast goal spaces, even generalizing to new tasks!
📄Paper: arxiv.org/abs/2502.07709
With 🧭MAGELLAN, our agent predicts its own learning progress across vast goal spaces, even generalizing to new tasks!
📄Paper: arxiv.org/abs/2502.07709
🚀 Introducing 🧭MAGELLAN—our new metacognitive framework for LLM agents! It predicts its own learning progress (LP) in vast natural language goal spaces, enabling efficient exploration of complex domains.🌍✨Learn more: 🔗 arxiv.org/abs/2502.07709 #OpenEndedLearning #LLM #RL
MAGELLAN: Metacognitive predictions of learning progress guide...
Open-ended learning agents must efficiently prioritize goals in vast possibility spaces, focusing on those that maximize learning progress (LP). When such autotelic exploration is achieved by LLM...
arxiv.org
March 24, 2025 at 7:53 PM
Humans excel at estimating their own competence and progress on tasks. But can LLM agents do the same?
With 🧭MAGELLAN, our agent predicts its own learning progress across vast goal spaces, even generalizing to new tasks!
📄Paper: arxiv.org/abs/2502.07709
With 🧭MAGELLAN, our agent predicts its own learning progress across vast goal spaces, even generalizing to new tasks!
📄Paper: arxiv.org/abs/2502.07709
Enabling forms of metacognition in LLMs is a frontiers challenge in #AI
We've made progress in this direction with MAGELLAN 🧭: curiosity-driven LLMs learn to predict and generalize their own learning progress in very large spaces of goals 🚀
This is then used for automatic curriculum learning 👇
We've made progress in this direction with MAGELLAN 🧭: curiosity-driven LLMs learn to predict and generalize their own learning progress in very large spaces of goals 🚀
This is then used for automatic curriculum learning 👇
🚀 Introducing 🧭MAGELLAN—our new metacognitive framework for LLM agents! It predicts its own learning progress (LP) in vast natural language goal spaces, enabling efficient exploration of complex domains.🌍✨Learn more: 🔗 arxiv.org/abs/2502.07709 #OpenEndedLearning #LLM #RL
MAGELLAN: Metacognitive predictions of learning progress guide...
Open-ended learning agents must efficiently prioritize goals in vast possibility spaces, focusing on those that maximize learning progress (LP). When such autotelic exploration is achieved by LLM...
arxiv.org
March 24, 2025 at 5:04 PM
Enabling forms of metacognition in LLMs is a frontiers challenge in #AI
We've made progress in this direction with MAGELLAN 🧭: curiosity-driven LLMs learn to predict and generalize their own learning progress in very large spaces of goals 🚀
This is then used for automatic curriculum learning 👇
We've made progress in this direction with MAGELLAN 🧭: curiosity-driven LLMs learn to predict and generalize their own learning progress in very large spaces of goals 🚀
This is then used for automatic curriculum learning 👇
Reposted by Pierre-Yves Oudeyer
“The days of pretending we have binary, easy-to-recognize categories of machines & living beings are over.”
— @drmichaellevin.bsky.social
— @drmichaellevin.bsky.social
Living Things Are Not Machines (Also, They Totally Are) | NOEMA
Our formal models of life, computers and materials fail to tell the entire story of their capabilities and limitations.
www.noemamag.com
March 18, 2025 at 5:25 PM
“The days of pretending we have binary, easy-to-recognize categories of machines & living beings are over.”
— @drmichaellevin.bsky.social
— @drmichaellevin.bsky.social
Reposted by Pierre-Yves Oudeyer
From the point of view of science I've never been too excited about the Game of Life. But now I realize that what's really exciting about it is what it tells us about what we can call metaengineering...
writings.stephenwolfram.com/2025/03/what...
writings.stephenwolfram.com/2025/03/what...
March 18, 2025 at 6:35 PM
From the point of view of science I've never been too excited about the Game of Life. But now I realize that what's really exciting about it is what it tells us about what we can call metaengineering...
writings.stephenwolfram.com/2025/03/what...
writings.stephenwolfram.com/2025/03/what...
Reposted by Pierre-Yves Oudeyer
New research project: Lluminate - an evolutionary algorithm that helps LLMs break free from generating predictable, similar outputs. Combining evolutionary principles with creative thinking strategies can illuminate the space of possibilities.
www.joelsimon.net/lluminate
www.joelsimon.net/lluminate
March 12, 2025 at 7:04 PM
New research project: Lluminate - an evolutionary algorithm that helps LLMs break free from generating predictable, similar outputs. Combining evolutionary principles with creative thinking strategies can illuminate the space of possibilities.
www.joelsimon.net/lluminate
www.joelsimon.net/lluminate
Very good points Thomas Wolf ! This is to address this grand challenge that various labs (and mine in particular) have been working on mechanisms enabling AI systems to have capabilities for curiosity and metacognition 🧵
I shared a controversial take the other day at an event and I decided to write it down in a longer format: I’m afraid AI won't give us a "compressed 21st century"
Here: thomwolf.io/blog/scienti...
It's an extension of this interview discussion from the AI summit: youtu.be/AxBd3G0lFLs?...
Here: thomwolf.io/blog/scienti...
It's an extension of this interview discussion from the AI summit: youtu.be/AxBd3G0lFLs?...
March 7, 2025 at 11:23 AM
Very good points Thomas Wolf ! This is to address this grand challenge that various labs (and mine in particular) have been working on mechanisms enabling AI systems to have capabilities for curiosity and metacognition 🧵