The Archive

Q-GAIN Python package applies ML and physics-informed methods to cold-atom Bose-Einstein condensate image analysis.

SPG-Layout generates physically plausible indoor scenes from text in non-orthogonal environments using LLMs and statistical priors.

Object-centric LeJEPA improves self-supervised vision representation learning by partitioning scenes at object level with reduced data requirements.

ACID enforces action consistency in world-model planning via inverse dynamics to ensure intermediate trajectory realizability during embodied control.

RFM-AGOP efficiently extracts multi-dimensional refusal subspaces in LLMs for safety steering, handling long reasoning traces computationally.

WattGPU predicts GPU power draw and inter-token latency for LLM inference across unseen hardware using only public specs, enabling deployment optimization.

DecompRL teaches LLMs modular code generation via RL to solve problems with near-zero base policy success rate, scaling accuracy beyond sampling limits.

Constraint-based oversight for coding agents scales cheaper than scaffolding; access control and tooling principles from human teams transfer directly.

Agentic search system deployed for NASA Earth Observation datasets; amplifies knowledge graph latent value via LLM-driven natural-language discovery.

TGO-II framework analyzes geometric evolution of Vision Transformer representations during supervised training via representational similarity.

Source-critical reasoning framework for RAG-based fact-checking; adds media background checks to handle conflicting, outdated, or biased evidence.

Multi-agent workflow (Gemini 2.5 Flash, RigoChat-7B-v2) for Spanish Easy-to-Read translation via LangGraph with event-condition-action routing.

Hardware-enforced semantic coordination for safety-critical autonomous systems; addresses bounded latency and deterministic multi-component orchestration.

DRIFTLENS measures reasoning drift in personalized LLMs; memory injection reshapes inference trajectories on open-ended questions without ground truth.

VisionAId: offline-first Android multimodal assistant for visual impairment; six on-device deep learning models for real-time object detection and personalization.

Agent-based patching of LLVM compiler missed optimizations; agents struggle with generalization beyond single cases, compared via benchmark.

Essay connecting literary analysis (narratology, translation, critical theory) to culturally literate AI development; plural LLM interpretations.

Study shows LLM personas have frame-dependent geometric structure distinct from aggregate Big Five traits using GPT-4o on psychometric questionnaires.

Proposes bounded memory gates for quantum fast-weight programmers to prevent divergence in long-sequence modeling tasks.

Geometry-aware attention framework GAP-GDRNet for 6D spacecraft pose estimation from monocular images in rendezvous scenarios.

SkillFuzz detects implicit intents when composed LLM agent skills in marketplaces redirect execution toward unintended objectives.

Self-gating attention mechanism reduces quadratic complexity of transformers in time series forecasting via redundant attention map pruning.

SelectTSL combines sound source localization with multimodal prompts for selective target sound direction estimation in complex acoustic scenes.

Certify-then-Rectify framework adds correctness guarantees to HNSW vector search via distribution-free statistical certification without sacrificing speed.

Frameworks like Lean Six Sigma and business process management (BPM) first gained traction because they promised clarity in the chaos—a structured way to bring order to messy, sprawling operations. Lean Six Sigma emphasized statistical rigor and quality control; BPM created end-to-end maps of how work should flow across departments. Both offered a repeatable way to…

Pipeline automating physics research from arXiv corpus to publication via LLM agents with external literature grounding to reduce hallucination.

OpenAI CEO Sam Altman has reportedly proposed giving 5% of the company’s equity to a U.S. sovereign wealth fund, reviving discussions about letting the public share in the financial gains from the AI boom.

CCG-based parser with directional type system improves structural generalization on SLOG benchmark, outperforming AM-Parser baseline.

HOLA augments linear-attention models with bounded exact KV cache inspired by complementary learning systems to improve long-range needle recall.

Insiders say Sam Altman is in active talks with the Trump administration.