The Relational Bottleneck as an Inductive Bias for Efficient Abstraction

N/A

The Relational Bottleneck as an Inductive Bias for Efficient Abstraction Taylor W. Webb1,* , Steven M. Frankland2 , Awni Altabaa3 , Simon Segert4 , Kamesh Krishnamurthy4 , Declan Campbell4 , Jacob Russin5 , Tyler Giallanza4 , Zack Dulberg4 , Randall O’Reilly6 , John Lafferty3 , and Jonathan D. Cohen4 1…

related reading

• Aman's AI Journal • Primers • Ilya Sutskever's Top 30aman.ai
• On the Learnability of Physical Concepts: Can a Neural Network Understand What’s Real?arxiv.org
• Natural Abstractions: Key Claims, Theorems, and Critiques — LessWronglesswrong.com
• What If We Had Bigger Brains? Imagining Minds beyond Ours-Stephen Wolfram Writingswritings.stephenwolfram.com
• The Natural Abstraction Hypothesis: Implications and Evidence — LessWronglesswrong.com
• 2023 letter | Zhengdongzhengdongwang.com
• [2508.11214] How Causal Abstraction Underpins Computational Explanationarxiv.org
• pdfopenreview.net
• Building machines that learn and think like people | Behavioral and Brain Sciences | Cambridge Corecambridge.org
• The Unbearable Slowness of Beingarxiv.org
• How Causal Abstraction Underpins Computational Explanationarxiv.org
• A critique of pure learning and what artificial neural networks can learn from animal brains | Nature Communicationsnature.com