https://news.ycombinator.com/item?id=45256179 :

> Which statistical models disclaim that their output is insignificant if used with non-independent features? Naieve Bayes [...]

Ironic then, because if transformers are Bayesian networks then we're using Bayesian networks for non-independent features.

From "Quantum Bayes' rule and Petz transpose map from the minimum change principle" (2025) https://news.ycombinator.com/item?id=45074143 :

> Petz recovery map: https://en.wikipedia.org/wiki/Petz_recovery_map :

> In quantum information theory, a mix of quantum mechanics and information theory, the Petz recovery map can be thought of as a quantum analog of Bayes' theorem

But there aren't yet enough qubits for quantum LLMs: https://news.ycombinator.com/item?id=47203219#47250262

"Transformer is a holographic associative memory" (2025) https://news.ycombinator.com/item?id=43028710#43029899