Neural Networks & Reasoning
Pete discusses the structural limitations of neural networks, emphasizing the challenge of modifying them without retraining. Daniel highlights recent advancements in interpretability and the ability to intervene in neural network behavior. Both reflect on the nature of language tasks, pointing out that while large language models can perform specific tasks, they lack a comprehensive world model, which limits their inferential reasoning capabilities.In this clip
From this podcast
The Gradient
Pete Wolfendale: The Revenge of Reason
Related Questions