Limitations of scaling LLMs and tokenization

Jakub Tomczak on: Why neurosymbolic approaches and domain knowledge could reshape the AI race. Is Europe poised to lead this new era?

In this episode, we dive into the resurgence of world models in AI and explore why the industry is revisiting neurosymbolic approaches. We challenge the dominance of data-centric models and discuss how combining symbolic knowledge with machine learning could unlock new breakthroughs. Our conversation highlights Europe's unique strengths and potential as a leader in this evolving landscape. We also address the importance of domain expertise and mathematics, and question whether the current AI race is stifling real innovation. Join us for a candid discussion on what the next wave of AI could—and should—look like.

We thank our partner SIEMENS

Siemens

https://new.siemens.com/global/en/products/automation/topic-areas/artificial-intelligence.html

OpenAI

https://openai.com/

Anthropic

https://www.anthropic.com/

Claude

https://www.anthropic.com/claude

Jürgen Schmidhuber

https://people.idsia.ch/~juergen/

UC Berkeley

https://bair.berkeley.edu/

NextAI Austria

https://www.nextai.at/

ME-AI

https://me-ai.org/

AWS (Amazon Web Services)

https://aws.amazon.com/

Apple

https://machinelearning.apple.com/

Promato

https://www.promato.com/

Jeff Bezos

https://en.wikipedia.org/wiki/Jeff_Bezos

Residual Neural Networks (ResNets)

https://en.wikipedia.org/wiki/Residualneuralnetwork

Kaiming He

https://scholar.google.com/citations?user=Phg3rFYAAAAJ&hl=en

ICML (International Conference on Machine Learning)

https://icml.cc/

Digital Twins

https://en.wikipedia.org/wiki/Digital_twin