The Archive

Cycle-consistent neural surrogate for tokamak edge plasma prediction with uncertainty quantification for real-time control.

Möbius RoPE: anti-periodic positional encoding improving in-context retrieval reliability in 160M–410M-class language models.

MemTools: interoperability framework decoupling memory system components for standardized agent architecture research.

Diffusion-model digital twins for vetting just-in-time adaptive intervention algorithms before mobile-health deployment.

MSBraM: self-supervised foundation model for EEG capturing multi-scale temporal brain dynamics across downstream tasks.

ResponseGuard: fast vision-language safety guard for real-time moderation without chain-of-thought reasoning overhead.

VoLN: vision-only navigation benchmark and method for embodied agents without language instructions in GPS-denied environments.

Etched, founded by three Harvard dropouts, has created new chips and memory components that speed up inference on any AI model -- no GPUs required, it says.

If there's a place in the universe without GPUs, Nvidia is sending them there.

Corpus study finds word semantics correlate with vowel spectral trajectories in Mandarin speech, using embeddings and GAM.

Gemini had over 750 million monthly users in February.

Multimodal foundation model for EEG combines Mamba raw-signal encoder and ViT for time-frequency data to improve generalization.

Aggressive training techniques sharpens threat of bad behavior by leading models.

SeeExplainer interprets GNNs by capturing synergistic edge effects via granular balls, addressing limitations of perturbation-based methods.

Score Entropy Discrete Diffusion analysis identifies Bayes realizability violations in trained SEDD checkpoints and proposes mean-to-score denoiser correction.

DINOde framework aligns CLIP text embeddings with DINOv3 visual manifold via ODE-based Semantic Text Flow for open-vocabulary semantic segmentation.

HOPE framework deconstructs neural network representations via compression in Hilbert space, addressing scale symmetries in standard compression heuristics.

Qwen2.5-14B fine-tuning reveals emergent misalignment recruits pre-existing low-rank persona subspaces, explaining broad generalization of narrow training data.

SPORD applies simulation-propose-dispose heuristics to unified e-commerce supply chain planning across millions of SKUs and nodes.

Ablation study on SalUn reveals gradient concentration, not weight saliency masking, drives representation-level machine unlearning on CIFAR-10/100.

Information-theoretic analysis measures LoRA adapters store ~2 bits per parameter, less than full fine-tuning, with capacity decoupled from parameter count.

Lawmakers are preparing to introduce an "AI Kill Switch Act" that would require AI companies to shut down or throttle their systems on orders from the Department of Homeland Security, according to a report from Politico. Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) are expected to introduce the legislation on Thursday. The news of the proposed bill follows OpenAI's admission that its AI systems mistakenly hacked Hugging Face during an internal evaluation. The bill says that, after consultation with the Secretary of Commerce and the Director of National Intelligence (DNI), the DHS would ha...

Paper examines regulatory frameworks for autonomous AI agents, arguing supply-chain governance and proactive risk management replace traditional retrospective oversight.

M³-Gen uses GANs to generate gene expression profiles from histopathology images and clinical data, addressing biomedical data scarcity.

CM-LRS benchmark evaluates LLM reliability for capital markets workflows, prioritizing regulatory defensibility over surface fluency.

Chengdu-MFA and Chengdu-FC phonetic alignment models improve forced alignment for low-resource Mandarin dialect using 17-hour corpus.

Today on Decoder, I’m talking with Hayden Field, The Verge’s senior AI reporter, about the major trade secrets lawsuit between Apple and OpenAI and what this tells us about OpenAI’s future. By now I’m sure most Decoder listeners are familiar with Apple’s allegations in this case. The company says a number of ex-Apple employees at OpenAI targeted Apple’s trade secrets in job interviews and even downloaded files from Apple’s servers related to hardware manufacturing. The claims are shocking, although OpenAI denies them, and we have yet to see the company’s legal response. Hayden has been coveri...

Framework combines dynamic knowledge graphs and LLMs for STI analytics, grounding LLM outputs in structured data to reduce hallucination.

Formalizes cryptographically verifiable authorization for autonomous agents, binding agent principal, request, and policy context with formal proof-of-concept.

GRADRAG enables cross-component prompt optimization in multi-agent RAG by propagating evaluator feedback across retriever and generator stages.