Archive

Week of September 7, 2026

2026-W37

Week 2026-W37: 20 curated papers with 20 free plain-English explainers.

93
#1
Artificial Intelligence2609.04194

Legibility is Not Interpretability: Comparing Judged and Actual Importance in Chain-Of-Thought Reasoning

Readable chain-of-thought traces look like a window into why a model answered as it did — but this paper shows that even strong LLM judges cannot reliably tell…

Read free explainer →

85
#2
Quantum Computing2609.04121

Quantum thermalization achieves optimal approximate quantum error correction

Generic thermalizing many-body dynamics, treated as an encoding, appear to hit the same optimal approximate error-correction limits as Haar-random codes — incl…

Read free explainer →

83
#3
Artificial Intelligence2609.04196

Puffin-World: Scaling a Unified Multimodal Model with Native 3D World States

Puffin-World is a single multimodal model that jointly handles physics, geometry, and appearance — plus camera motion — to generate and reconstruct 3D worlds w…

Read free explainer →

82
#4
Robotics2609.04193

GIFT: Guided Intermediate Feature Training via Action-Oriented Structural Supervision for Robotic Manipulation

GIFT trains a robot policy’s intermediate features to keep geometry, affordances, and goals — the structure that actually matters for control — and shows consi…

Read free explainer →

80
#5
Artificial Intelligence2609.04202

TokenMatch: 3D Mesh Correspondence Transformer with Curvature-Guided Tokenisation

TokenMatch is a feed-forward transformer that matches 3D meshes — including hard partial, non-isometric cases — by tokenizing surfaces with curvature-guided pa…

Read free explainer →

78
#6
Quantum Computing2609.04165

Parameterised graph theory for tensor networks: entanglement rerouting, structural simplification, and agnostic tomography

Graph parameters such as cutwidth and tree-cutwidth are shown to control how expensive it is to rewrite a tensor-network state as an MPS or tree network — and…

Read free explainer →

77
#7
Biotech & Longevity2609.02344

Subcellularly Resolved Single-Cell Embedding Learning with Transcriptomic data, Protein Structure and Localization Information

A multimodal cell-embedding model mixes RNA profiles with protein sequence and structure, using cross-attention inside subcellular compartments instead of trea…

Read free explainer →

75
#8
Advanced Materials2609.03795

Exciton multipolarity controls coherent and squeezed phonons in van der Waals heterostructures

In MoSe2/WSe2 multilayers, dipolar excitons drive coherent interlayer breathing phonons while quadrupolar excitons drive squeezed ones — and a vertical electri…

Read free explainer →

74
#9
Robotics2609.03984

MulDP: Multimodal Diffusion Policy for Autonomous Quadruped Parkour Navigation across Complex Terrains

MulDP is a diffusion policy that turns vision, proprioception, and goals into anticipatory velocity commands so a quadruped can parkour-navigate without a huma…

Read free explainer →

72
#10
Advanced Materials2609.03757

Ion-Engineered Insulator-to-Semiconductor Transition in Natural 2D Biotite

A controlled NaOH treatment turns exfoliated biotite nanosheets from an insulator into a tunable 2D semiconductor by adding sodium, defects, and local reconstr…

Read free explainer →

71
#11
Quantum Computing2609.04132

Characterizing Large Scale Quantum Systems with Error Per Circuit Layer

Error Per Circuit Layer (EPCL) estimates how polarizing each layer of a quantum circuit is by overlapping two registers that ran the same random circuits — no…

Read free explainer →

70
#12
Robotics2609.04147

A Low-Cost, Open Platform for End-to-End Autonomous Driving on a Miniature Ackermann Vehicle

An open miniature-car testbed — physical Ackermann vehicle, printed urban track, tools, and a Webots twin — lets researchers take end-to-end driving from simul…

Read free explainer →

68
#13
Biotech & Longevity2609.01357

PopPert: Population-level Joint-Distribution Modeling for Single-Cell Perturbation Prediction

PopPert predicts how a whole population of cells changes after a genetic or chemical perturbation by learning joint gene-expression distributions — without pre…

Read free explainer →

66
#14
Climate Tech2609.03210

Improving precipitation forecasts in an AI weather model using observational data

Fine-tuning a graph-transformer weather model on IMERG precipitation — not just ERA5 — improves medium-range rain scores by up to 19% and extreme-rain skill by…

Read free explainer →

65
#15
Artificial Intelligence2609.04189

Robust PAC Learning of Concurrent Stochastic Games

A PAC algorithm for general-sum concurrent stochastic games with uncertain transitions either returns a near-optimal approximate Nash equilibrium or certifies…

Read free explainer →

63
#16
Biotech & Longevity2609.02985

Sparse concept attribution for histomorphological hypothesis generation from whole-slide classifiers

SCOPE explains whole-slide pathology classifiers by sparsely attributing them onto a histomorphological concept bank — and a new benchmark shows sparse attribu…

Read free explainer →

59
#17
Space Systems2609.03975

Detection of hydrocarbons in Titan using high-resolution cross-correlation spectroscopy

Using Titan as a testbed, the authors build high-resolution cross-correlation templates from molecular cross-sections instead of complete line lists — recoveri…

Read free explainer →

57
#18
Space Systems2609.03779

Morphology of Radio Sources in Representation Space

A classifier ensemble plus representation-space clustering finds six extra radio morphologies in LoTSS-DR3 — including rare winged and FR-ambiguous sources — o…

Read free explainer →

55
#19
Climate Tech2609.02614

Ruelle--Pollicott Theory for Metastable Systems: A Unified Framework for Tipping Transitions

A Ruelle–Pollicott spectral theory of stochastic metastable systems shows that classic “critical slowing down” warnings are not universal: large responses need…

Read free explainer →

49
#20
Energy & Fusion2609.03768

Computing high-order mixed derivatives in physics-informed neural networks using multi-index Bell polynomials

A tabulated Faà di Bruno / Bell-polynomial recursion computes the mixed derivatives PINNs need — and their parameter gradients — without nested automatic diffe…

Read free explainer →