The Archive

HeRo enables selective disclosure watermarking in LLMs, allowing verification of watermark subsets without exposing full embedded metadata.

First multiplayer world model for dynamic environments attributes scene changes to multiple agents' actions, trained on 10K hours of Rocket League gameplay.

OptiAgent is a multi-agent framework that converts natural language OR problems to mathematical formulations and executable code via iterative self-correction.

TREK uses distillation for exploration support expansion in GRPO, enabling reasoning on hard prompts via black-box or white-box teacher trajectories.

Siamese networks require threshold selection for embedding-space verification; paper proposes unsupervised threshold definition without labeled data.

PSA: A change to Google's privacy settings let it train its AI on more of your data. Here's how to opt out.

Diffusion autoencoders exhibit distinct optimization regimes during training; study links reconstruction vs. disentanglement trajectories to early dynamics.

Medical imaging CNN struggles with aneurysm-bifurcation confusion; topology-aware framework using Smooth Euler Characteristic Transform reduces false positives.

Linear probes decode remaining output length from LLM hidden states across 7-8B open-weight models, revealing internal response-length estimation.

M3Bench evaluates model editing for medical vision-language models under clinical variation; tests reliability, precision, generalization post-deployment.

Quantum anomaly detection via spectral methods; computes quantum state deviation from normal dataset without explicit eigenvector recovery.

Pathway Activity Autoencoders integrate multi-omics data with biological priors for cancer risk stratification; balances interpretability and capacity.

Anthropic accused of spying on users; engineer says “experiment” is over.

Subspace-constrained LoRA adaptation on Flan-T5 resists poisoning; restricts updates to 128 coordinates in shared adapter subspace, blocking malicious fine-tuning.

MetaSkill-Evolve enables recursive self-improvement of LLM agents; agents rewrite and evolve their own improvement procedures from execution traces.

PM2.5 super-resolution downscaling over Europe using station-guided pseudo-supervision; bridges coarse atmospheric model fields to local ground observations.

Wavelet scattering transform applied to EEG for schizophrenia biomarker discovery; medical imaging application, not frontier AI.

Quantum hardware fingerprinting via noisy output distributions raises privacy concerns for cloud quantum services.

Neural Controlled Differential Equations for continuous-time modeling of irregular time series; incremental method advance.

Defense mechanism for web agents against prompt injection via content masking with formal security guarantees.

ProPS framework generates speaker embedding distributions conditioned on natural language descriptions; specialized speech task.

Transfer learning method for physics-informed neural networks in PDE inverse problems; domain-specific technical advance.

RL-based adaptive batching and routing for inference serving outperforms static policies on production traces.

Audit of hyperbolic geometry in vision-language models identifies operating-point failure modes; questions practical utility.

Continuous reference biomass data from QSM improves LiDAR-based forest AGB estimation; specialized remote sensing application.

SalAngaBhava: aspect-based sentiment analysis dataset for Sinhala (low-resource language); dataset contribution.

GeoFlow augments graph neural networks with geospatial attributes for origin-destination flow prediction in urban mobility.

Government of Alberta deploys Claude Opus and Sonnet via Claude Code for cybersecurity vulnerability detection and remediation across government systems since 2025.

I’m just doing my job, but turns out, it feels lousy holding other people’s privacy in your hands. I've long argued that Hollywood has simultaneously set and ruined our expectations for smart glasses. But after binge-watching two seasons of Netflix's A Man on the Inside, this is perhaps the first time I've seen Hollywood, perhaps inadvertently, illustrate the biggest cultural problem with smart glasses as they stand today. In a nutshell, Ted Danson plays Charles Nieuwendyk, an elderly widower who finds a new purpose working for a private investigator. Armed with a pair of Ray-Ban Meta glasses...

FUSE applies flow matching with Feynman-Kac steering for multimodal simulation-based posterior inference.