The Archive

Current AI launches Gap Map v0.1, an index of 421 open-source AI products across models, tools, datasets, and hardware, backed by $400M committed capital.

Developer Josh W. Comeau reports significantly lower course sales, attributing decline to AI-driven job displacement fears and LLM-based tutoring competition.

NetinfoGC proposes permutation-invariant graph representations for classification using structural descriptors and training-free NUI estimation.

Modular foundation model framework for time-series perception in Digital Twins and PHM systems with improved transferability.

PLGSA-Transformer uses periocular landmarks and adaptive thresholding for masked face recognition across modalities.

Study of geometric perturbations via dihedral group transformations in UNet, ViT, and DiT architectures using activation diagnostics.

Efficient runtime representation for neuro-symbolic learning that decouples program search from parameter calibration via evaluator differentiation.

Knowledge distillation framed as learning teacher representation equivalence classes rather than absolute feature coordinates.

EPRA U-Net architecture for acute ischemic infarct segmentation in diffusion-weighted MRI with EfficientNet encoder.

Simon Willison reports Anthropic's Claude Code team recommends letting Fable/Opus apply autonomous judgment on tasks like testing rather than prescribing workflows.

Scalable AI safety via doubly-efficient interactive proofs that relax debate assumptions of equal model ability and truthfulness.

PULS pipeline for video anomaly anticipation using kinematic-to-semantic distillation bridge and V-JEPA 2 with Qwen3-VL embeddings.

WeightCLIP learns dataset-aligned latent space for neural network weights using contrastive learning on model and dataset embeddings.

Fuzzy logic extension of Answer Set Programming bridges vague qualitative reasoning with numerical data.

Systematic review of non-social-media datasets for mental health disorder detection via NLP and ML.

MentalThink equips multimodal LLMs with SVG-based visual reasoning for structured spatial problem-solving.

Selective prediction framework enables LLMs to abstain on uncertain inputs, balancing risk-coverage tradeoff.

GameEngineBench evaluates coding agents on C++ implementation tasks in Unreal Engine real-time environments.

Generator-fixer self-play with automatic curriculum scales code repair via RL without paired datasets.

Co-adaptive LoRA determines domain participation weights and multi-task grouping without labels via forward passes.

Theoretical analysis shows projected Adam with arbitrary moment decay exhibits non-zero average regret.

Mixture-of-Gaussians framework provides principled schedule design for Brownian Bridge Diffusion Models.

AGL-1 governance layer enforces authorization, audit, and agent constraints across distributed enterprise AI systems.

Two-stage retrieval system combining lightweight embeddings with expensive rankers for efficient candidate selection in similarity-based search.

AquaGen generates all-atom molecular configurations matching Boltzmann distributions faster than molecular dynamics simulations.

CAGE-1 benchmark evaluates enterprise agentic AI on authorization, policy compliance, tool safety, and memory validity in operational workflows.

Graph p-Laplacian solver for semi-supervised learning using continuation method to handle nonlinear energy minimization.

DeepSeek V3 and Kimi K2 perform structured reasoning over filler tokens with no visible output, revealing hidden computation interpretability challenges.

STRATOS framework bridges symbolic-numeric gap in Text-to-SQL querying for meteorological datasets via domain-specific engineering.

Comprehensive MI estimation benchmark suite using copula theory to evaluate estimators on high-dimensional, realistic distributions.