π New internship positions available in the Flowers AI & CogSci lab @Inria Bordeaux !!!
Inria Flowers team
@FlowersINRIA
FLOWing Epigenetic Robots and Systems, Inria research team on developmental and social robotics.
53 Following 1.7K Followers 624 Posts
Latest posts 5
Very important work and paper, congratulations @liweijianglw and co-authors !
β οΈDifferent models. Same thoughts.β οΈView quoted post on X
Todayβs AI models converge into an ππ«ππ’ππ’ππ’ππ₯ ππ’π―ππ¦π’π§π π, a striking case of mode collapse that persists even across heterogeneous ensembles.
Our #neurips2025 π&π ππ«ππ₯ π©ππ©ππ« (β¨ππ¨π© π.ππ%β¨) dives deep into this phenomenon, introducing ππ§ππ’π§π’ππ²-ππ‘ππ, a real-world dataset of 26K real-world open-ended user queries spanning 17 open-ended categories + 31K dense human annotations (ππ π’π§πππ©ππ§πππ§π ππ§π§π¨ππππ¨π«π¬ π©ππ« ππ±ππ¦π©π₯π) to push AIβs creative and discovery potential forward.
Now you can build your favorite models to be truly original, diverse, and impactful in the open-ended real world.
πPaper: arxiv.org/abs/2510.22954
πData: huggingface.co/collections/liβ¦
We also systematically reveal Artificial Hivemind across:
π₯ πππ§ππ«πππ’π―π πππ’π₯π’ππ’ππ¬: not only do individual LLMs repeat themselves, but different models produce strikingly similar content, even when asked fully open-ended questions.
π₯ ππ’π¬ππ«π’π¦π’π§πππ’π―π πππ’π₯π’ππ’ππ¬: LLMs, LM judges, and reward models are systematically miscalibrated when rating alternative responses to open-ended queries.
(1/N)
Hundreds of new generative AI models, some close, some open, are released every months.
Most of them derive in some way (weights, data, architecture,etc) from previous models, but this information is rarely disclosed.
Can one infer their ancestors? Yes, with PhyloLM π
Most of them derive in some way (weights, data, architecture,etc) from previous models, but this information is rarely disclosed.
Can one infer their ancestors? Yes, with PhyloLM π
π₯Our paper PhyloLM got accepted at ICLR 2025 !π₯View quoted post on X
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 and @pyoudeyer ! Here is a brief recap β¬οΈ
Excited to share with @flowersINRIA ADTool, an open-source Python toolkit using curiosity-driven AI to explore behavior diversity of complex systems (cellular automata, physics models, ...)!π
Serving off-the-shelf strategies like IMGEP
github.com/flowersteam/adβ¦
You said IMGEP? π
Serving off-the-shelf strategies like IMGEP
github.com/flowersteam/adβ¦
You said IMGEP? π
2:04
An overview of the work of the team @FlowersINRIA on grounding language with curiosity-driven agents (including LLM agents) !
πHow to ground generative AI in the external world?View quoted post on X
π€How to turn LLMs into curious agents that spontaneously explore, set their own goals, and become open-ended learners?
π€How to align such agents on the social environment?
πΊIn this talk, I outline a research program I have been developing with many colleagues over the last decade, and present experiments addressing these questions.
Thanks a lot to Stevan Harnad who invited me give this talk in his interdisciplinary seminar series (and stimulating questionsπ‘at the end of the talk!) π
Video link below