The Archive

LLMs evaluated as semantic evaluators for ASR, achieving 92-94% human agreement compared to traditional WER metrics.

Analysis demonstrates fine-tuning parameter subspace choice fundamentally alters continual learning problem definition and outcomes.

Theoretical result: multicalibration requires Θ(ε⁻³) sample complexity for achieving Expected Calibration Error bounds.

MathDuels benchmark uses self-play with LLMs as both problem authors and solvers to differentiate model capabilities beyond static benchmarks.

HalluScope benchmark reveals prompt-induced hallucinations in LVLMs stem primarily from language dominance over vision grounding.

Link: anthropic.com

Agentic architecture automates conversion of research questions to reproducible scientific workflows via LLM semantic translation and domain expert Skills.

User asks whether local models (32-128GB RAM) deliver genuine productivity or remain hobbyist tools.

Survey frames LoRA through signal processing lens, unifying architectural choices and optimization techniques for parameter-efficient fine-tuning.

Scale-adaptive diffusion framework enables joint spatiotemporal video super-resolution across variable upscaling factors and frame rates.

GiVA uses gradient-informed initialization to improve vector-based adaptation efficiency, matching LoRA training times with extreme parameter efficiency.

Computational framework analyzes Brazilian parliamentary discourse using stylometric analysis and topic modeling on legislative speech.

Nemobot is an interactive environment for creating and deploying LLM-powered game agents across multiple game classes using Claude Shannon's taxonomy.

Study incorporates environmental and visual predictors into motor insurance claim-frequency models using zone-level geographic data.

Multi-stage deep learning framework with warm-start optimization addresses Unit Commitment problem for grid scheduling with renewable integration.

EVENT5Ws is a large manually-annotated open-domain event extraction dataset covering diverse event types for improved algorithm development.

TingIS is an enterprise-scale system for real-time risk event discovery from noisy customer incident data in cloud-native services.

Multimodal text-graph approach leverages LLMs for open-domain event extraction without predefined event types using document-level context.

Claude introduces segmented rate limits: Claude Design separate quota, daily routine runs cap (0/15), and faster reset cycles; suggests potential billing/product changes.

RedirectQA dataset uses Wikipedia redirects to study LLM memorization of factual knowledge via entity surface form variation.

Study examines image authenticity risks from generative AI hallucinations in camera image signal processors (ISPs) at capture-time.

User observes API rate-limit reset time shifted from Thursday to Saturday; speculates new model launch.

Study evaluates whether LLMs encode relational context in moral decisions using Whistleblower's Dilemma across crime severity and social closeness dimensions.

User reports Claude rate limits were reset.

Machine learning model identifies mechanistic reasoning moments in student STEM team conversations via interpretable probability scoring.

Reddit user speculates that AI latency will decrease over time, comparing current Claude usage to dial-up internet speeds.

Figure AI shows progress on 03 humanoid robot production scaling and deployment.

ReaPER+ replay buffer optimization addresses quantum circuit learning under hardware noise via annealed TD error prioritization.

LLM sampling strategies for reasoning vs. output differ across languages; decoupling temperature affects both quality and determinism.

Transient Turn Injection attack exploits stateless moderation in commercial and open-source LLMs via multi-turn adversarial distribution.