Can AI learn better by explaining itself? 🧠🤖 Our paper 'Learning by Self-Explaining,' published in TMLR, explores how AI models improve generalization and avoid shortcuts by evaluating their explanations. Dive in for more: tinyurl.com/2z7v6kbt #AI #ML #ExplainableAI #TMLR

Very excited to be in Barcelona for NeSy 2024! We're presenting two papers: - Our ICML paper x.com/EmilevanKrieke… - A preview of our new NeSy library ULLER together with Samy Badreddine, Eleonora Giunchiglia and Robin Manhaeve. We will give a special tutorial on Thursday! Hit me up!

Excited to be at NeSy! ULLER is a project for better accessibility in Neurosymbolic research. Come to our talks for more details 🚀

#NeSy 2024 is on!! Live-streamed: sites.google.com/view/nesy2024 The community keeps growing! Artur d'Avila Garcez @[email protected] Pranava Madhyastha Roberto Confalonieri Ben Wagner Neurosymbolic AI @ City City STEM City, University of London Sony AI

Check out Samy Badreddine presenting our work on ULLER! We will share more soon :) If you're here in Barcelona, come join us Thursday morning for our tutorial & discussion session! Let's make NeSy accessible 🔥

Check out our work on tokenization led by the amazing Renato Lui Geh! It turns out you should consider *many* ways of tokenizing a sentence, which surprisingly gives rise to a #neurosymbolic problem formulation!

Happy with all the feedback from a packed room! We gave a tutorial of our upcoming ULLER NeSy library, and are excited to share more soon.

Today we presented our new ULLER library for Neuro-symbolic AI! We are very grateful for the engagement shown by the community and the feedback received! 💪

This is also close to my guess of how o1 is trained (my default frame is to view it as "inferring the best latent CoT that explains the data using a branching process like SMC", but of course the RL / MCTS frame would be more natural to folks at OpenAI).

martin_casado Seems to me Gary Marcus has been right about almost everything related to the limits of LLMs. We are now seeing systems with search, which is something all of us old timers have been waiting for the LLM/AGI crowd to rediscover. Maybe symbolic abstraction is next?

Unless you are an OpenAI employee working on improving their products, I don’t understand why such efforts are science. Why are we (question to faculty) spending taxpayer dollars in doing QA for a closed product by a well-capitalized company that does not give back to science?

Very nice to see Qwen ditching the mess that MMLU is, and using our reannotated MMLU-Redux for evaluation 🔥 arxiv.org/abs/2406.04127 huggingface.co/datasets/edinb…

RLHF is a popular method. It makes your human eval score better and Elo rating 🚀🚀. But really❓Your model might be “cheating” you! 😈😈 We show that LLMs can learn to mislead human evaluators via RLHF. 🧵below

🎊 It has arrived 🎊, the 2nd edition of my "Deep Generative Modeling" book. It has 100 new pages, 3 new chapters (incl. #LLMs) and new sections. It covers all deep generative models that constitute the core of all #GenerativeAI techs! Check it out: 💻tinyurl.com/mwj9dw83

What's better than waking up from a nap with an accept at #NeurIPS24? 😎 Well... for today is enough 😅 Thanks to all my coauthors, Samuele Bortolotti , Emile van Krieken , Tommaso Carraro, Paolo Morettin, antonio vergari - hiring PhD students, andrea passerini, Stefano Teso arxiv.org/abs/2406.10368

Happy to share that our papers on concept learning (tinyurl.com/3urm5uhb) and CBMs for RL (tinyurl.com/r8hm7r22) have been accepted at NeurIPS 2024! 🎉 Both explore ways to enhance AI interpretability and reasoning. Excited for the discussions #NeurIPS2024 #Interpretability

So happy that our paper Interpretable Concept Bottlenecks to Align Reinforcement Learning Agents (arxiv.org/pdf/2401.05821) has been accepted at NeurIPS 2024! 🎉 If you are wondering why RL agents cannot generalize to new scenarios and how to mitigate it, check it out !

Excited to share our #NeurIPS2024 work on combining diffusion models with constrained optimization to generate data adhering to constraints and physical principles (with guarantees!). Led by the amazing Jacob Christopher and with Stephen Baek. > arxiv.org/abs/2402.03559

Ordering merch for a minecraft mod I composed music for a decade ago is pretty mindblowing! Thanks The Aether for this initiative :)