The Archive

Finnish Baltic Sea AIS maritime dataset with 229M data points for collision prevention and cargo route analysis.

ARC proposes three-layer governance framework for autonomous robots: safety validation, cognitive certification, and operational authorization.

GPU utilization analysis reveals Nvidia Hopper's SM metrics hide inefficiency in LLM decode phase due to small-batch matrix fragmentation.

Python 3.15 soft-deprecates re.match() in favor of re.prefixmatch() for clarity.

PA-CDM introduces position-aware character detection matching for handwritten math expression recognition, improving position-blind metrics.

DualMLC dual-branch framework combines heterogeneous LMs for large-scale multi-label text classification with thousands of candidate labels.

Three-stage pipeline trains compact dense retrievers for Polish and European languages via cross-lingual alignment and knowledge distillation.

MTD3-BC offline RL algorithm optimizes wind farm yaw control under varying wind directions using precollected offline datasets.

ns-3 simulation estimates 802.11 contention time costs in federated learning by predicting frame-delivery ratios across client densities.

Physics-enriched neural solver for glacier ice-flow simulation warm-starts weights from low-order balance fields across time steps.

Physical-State-Guided Diffusion Sampling couples learned geological priors with seismic waveform solvers via Gaussian bridge for subsurface velocity reconstruction.

Meta says it's making changes to the prompts suggested by its AI chatbot after a viral video showed it digging for personal information about a woman's young daughters, as reported earlier by Futurism. In a statement to The Verge, Meta spokesperson Dina El-Kassaby says the company "missed the mark," adding that "the feature never should have prompted the individual with questions like that." Last week, Instagram user Kalie Robins posted a video explaining how Meta AI presented her with an invasive suggestion after cross-posting a clip to Facebook. The AI prompt, "Who is the child passenger?" ...

UniPart enables zero-shot language-grounded 3D part segmentation for robotic manipulation using cross-modal Transformer conditioning on CLIP embeddings and LangPart-1M dataset.

Study of cross-layer influence propagation in multimodal model merging reveals coupling effects between visual and textual expert updates across network depth.

Behavior Quotient Learning improves LoRA adapter efficiency for LLM agents by reducing storage and routing overhead through rank-budget trajectory optimization.

Reliability-Aware Boundary Refinement Network (RABR-Net) adds uncertainty-guided boundary refinement to biomedical image segmentation for reliable clinical contour delineation.

PAC-Bayesian framework quantifies knowledge transfer from privileged information available only at training time into generalized model performance bounds.

Internal Dual-Wiener routing mitigates gradient dominance in long-horizon autoregressive forecasting by reliability-weighting credit assignment across timesteps.

MedSNIP-Bench introduces snippet-level granularity for medical fact verification, preserving clinical structure through clause grouping instead of atom-level decomposition.

Audit of Pyrfume olfaction descriptor corpora reveals measurement inconsistencies across datasets with median tetrachoric agreement 0.795 vs. Cohen's κ 0.413.

DuplexDrama synthesizes 2000+ hours of dialogue audio with full-duplex behaviors, expressive emotion, personas, and sound events from 64-voice timbre pool.

Graham Dumpleton releases wrapture, a Python monkey-patching library for testing and observability tracing.

Some dangerous biology looks much like legitimate research, complicating AI safeguards.

OpenAI scaled Habitat storage platform from Python library to serve 1B ChatGPT users at 22M requests/sec globally.

Datasette security patches (1.0a39, 0.65.4) address vulnerabilities in public/private table isolation, audited with Claude and GPT models.

datasette-publish-fly 1.4 adds force_https, fixes volume bug, supports app-scoped deploy tokens.

github-to-sqlite 2.9.1 fixes sqlite-utils 4.x compatibility.

Datasette 1.0a39 security release addresses vulnerabilities.

trynix.dev runs Nix packages in a browser-based QEMU-WASM VM, enabling reproducible environment access via URL for 13 years of package history.

Nvidia has its finger in every pie, and sees another year of plenty in its future, Jensen Huang says. But, he insists, its deals are not circular.