This paper from Meta about "Soft Tokens" in RL is interesting; it allows LLMs to invent their own non-discrete (recursive) representations in order to solve problems better… Results are mixed though: it's only a few percent better on GSM8k from pass@4 onwards, and pass@32 just
MACHINE LEARNING
-
Deploying Machine Learning Models on Arduino Microcontrollers
By
–
From Blink to Think: Deploying ML on Arduino! At LearnOpenCV, we’ve always believed that AI shouldn’t be limited to powerful GPUs or cloud servers. It should run everywhere – even on the tiniest boards.
Our latest article of the edge devices series, explores exactly that idea. -
Neural Operators and TensorLy Ecosystem at PyTorch Conference
By
–
It was wonderful to have @JeanKossaifi and David present the poster on Neural Operators and tensorly at @PyTorch yesterday. There is no ai+science without neural operators. Check out our rich ecosystem.
-
New AI Model for Drug Design Integrates Advanced Physics Predictions
By
–
Proud to have contributed to this https://
caltech.edu/about/news/new
-ai-model-for-drug-design-brings-more-physics-to-bear-in-predictions
… -
PyTorch Conference: Neural Operators for AI Science
By
–
My first @PyTorch keynote was almost a decade ago in a small venue. So much has changed. Was amazing to come back to the much bigger @PyTorch conference and talk about ai+science. Check out our Neural Operator library and its latest release https://t.co/CouRPFdF60 https://t.co/bOniHIvZlZ
— Prof. Anima Anandkumar (@AnimaAnandkumar) 23 octobre 2025My first @PyTorch keynote was almost a decade ago in a small venue. So much has changed. Was amazing to come back to the much bigger @PyTorch conference and talk about ai+science. Check out our Neural Operator library and its latest release https://
github.com/neuraloperator
/neuraloperator
… -
Neural Operators: The Future of AI and Science
By
–
I will be giving a keynote in 30 minutes on foundations for ai+science and @JeanKossaifi will be presenting a poster on the neural operator library. Check it out! The future of ai+science cannot be conceived without neural operators.
-
PyTorch’s Unexpected Growth While Preserving Core Values
By
–
. @PyTorch has snowballed into something never imagined, while still keeping our core values intact!
feels incredible.
-with @ezyang and @apaszke -
VISTA AI Model Improves Performance Through Self-Learning Feedback
By
–
VISTA improves without ever retraining the model. It uses feedback from its own outputs to rewrite future prompts like an AI director learning with every take. The results? 60% pairwise win rate vs state-of-the-art
66.4% preference by human evaluators -
PyTorch Framework Continues Growth While Maintaining Core Values
By
–
. @PyTorch has snowballed into something never imagined, while still keeping our core values intact!
feels incredible.
-with @ezyang and @apaszke -
Datalab’s State-of-the-Art Text Extraction Models Now Available
By
–
Datalab's models are state-of-the-art for text extraction, and are based on the popular open-source Marker and Surya projects. They're fast and cost between $2-$6 per 1000 pages.