On the Dynamics of Local Hidden-Variable Models
Shows that quantum correlations can remain Bell-local at every instant yet resist local hidden-variable dynamics in interacting systems, unlike noninteracting ones.
https://t.co/q8eH37HWQ6
Why Finite Intelligence Works in an Infinite World: A Mathematical Theory
Defines the information needed to preserve decision quality across future tasks and derives when finite representations suffice and how shared task structure sets their size.
https://t.co/Fa0znbG5lt
Guerra Interpolation and Optimal Transport: Kantorovich Bounds, Brenier Maps and the Replicon
Connects spin-glass interpolation and the Parisi formula to optimal transport, and finds the replicon in a path that splits a point mass.
https://t.co/ZHe27QrP0e
Incentive-Responsive Agents: A Principal-Agent Framework, Metric Suite, and Runtime Architecture for Long-Lived Personal AI Agents
A principal-agent model and experiments show runtime allocation boosts success, unlike incentive prompts.
https://t.co/y1q3aPSVKs
A reason why we do not observe Schr\"odinger's cats
Under an ensemble assumption, a stochastic equation derived from Schrödinger dynamics predicts rapid collapse of macroscopic superpositions with Born-rule probabilities.
https://t.co/kRFToSybfu
A Diophantine definition of $\mathbb{Z}$ in $\mathbb{Q}$
A degree-four polynomial defines the integers within the rationals, making recursively enumerable rational sets Diophantine and refuting Mazur's topology conjecture.
https://t.co/HCgkqGjYkb
Scaling limit of the Random Language Model
A large-grammar scaling theory predicts a freezing transition, distinct fluctuation regimes, and Heaps' and Zipf's laws.
https://t.co/eXfhJvedKw
Inverse Scattering & Isospectrality in (Analog) Cosmology
Expansion and contraction produce identical particle numbers, but adding phase information allows unique reconstruction of regular, asymptotically static cosmological histories.
https://t.co/Td6mH4ECfk
Recursive memory for prediction and generation
Predicting a hidden Markov source takes O(log(1/ε)) bits, but self-fed T-token generation can need another log T bits; maintaining sliding-window counts shows a related gap.
https://t.co/gjNWezbE7Z
Heat as gauge
A gauge-theoretic formulation treats heat as a connection, derives a geometric Second Law from curvature, and identifies an entropy-shift analogue of the Aharonov–Bohm effect.
https://t.co/FrTzFz0khW
You Don't Need Feature Learning to Do In-Context Learning
A spectral analysis shows kernel machines first generalize and then memorize in-context regression as data grows, with transformers exhibiting the predicted memorization.
https://t.co/E60HQwNA9d
Prestatability in the Macroscopic Limit
Shows that future macroscopic variables may be impossible to specify fully in advance, with their relevance undecidable under universal dynamics.
https://t.co/c2QwfM1I3C
Sparsity and Out-of-Distribution Generalization
Proves a generalization bound for sparse classifiers when training and test distributions overlap on relevant features, even if other features shift.
https://t.co/pBeFeNa4nz
The Lagrangian Brain: A Constraint Mechanics of Mind and Machine
Proposes constraint mechanics for neural circuits and reports a gradient-free machine that solves 422,786 extreme Sudoku puzzles and generalizes to larger mazes and longer strings.
https://t.co/WFQkkf0erG
A calculus of types in Isbell nuclei
Identifies orthogonality-generated logical types with fixed points of an Isbell adjunction and derives a noncommutative Lambek calculus.
https://t.co/r9eE54nunl
Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data
Proves that outcome-only RL teaches linear transformers stepwise graph traversal when training includes enough simple examples.
https://t.co/fYsLs74QRj
Against All Likelihood
Argues that truth is better gauged by invariance across relevant environmental transformations than by likelihood, and sketches tests for language models.
https://t.co/QyPAVGcVIp
The Logic of Emergence
Extending model theory, the authors formalize relationships among scientific theories across scales, finding multiple hierarchies and a possible collapse into one field.
https://t.co/aDool0ugws
How Finite Brains Generalize Without Bound: Spatial Memory as a Universal Computer
A neural model reads repeatable coincidences in cyclic memory rather than positions, enabling universal computation and long-string generalization.
https://t.co/YducF79GQW
Deep-layered machines have a built-in Occam's razor
An exactly solvable model shows that adding layers increasingly favors simple output functions, with probabilities approaching 2^-K for Kolmogorov complexity K.
https://t.co/LK54fbXd6i