The Theory of Contravariance, Part 0 - by Dan Yamins

Breaking down our extensive new theory paper.

The Theory of Contravariance, Part 0 Breaking down our extensive new theory paper. Dan Yamins Jul 13, 2026 9 Share Aran Nayebi and I have just released a rather, uhm … let’s say, sizeable… new paper on the Theory of Contravariance . Because it’s a long paper and has many different ideas in it, we’re going to be writing a multi-part substack series on it, beginning with this post describing the overall purpose of the project. What needs to be explained in NeuroAI? Why is there apparent convergent evolution between task-driven neural network models and the brain? Over the past 15 years in NeuroA

related reading

• Alignment remains a hard, unsolved problem — LessWronglesswrong.com
• The Platonic Representation Hypothesisphillipi.github.io
• Naturally Occurring Equivariance in Neural Networksdistill.pub
• Neuroscience is pre-paradigmatic. Consciousness is whytheintrinsicperspective.com
• The hot mess theory of AI misalignment: More intelligent agents behave less coherently | Jascha’s blogsohl-dickstein.github.io
• The Platonic Representation Hypothesisarxiv.org
• Continuous Thought Machinespub.sakana.ai
• Ngo and Yudkowsky on alignment difficulty — LessWronglesswrong.com
• Perfectly Normalperfectlynormal.co.uk
• A deep learning framework for neuroscience - PMCncbi.nlm.nih.gov
• Human-like Neural Nets by Catapulting · Gwern.netgwern.net
• Formation of Representations in Neural Networksarxiv.org