The Archive

EDGE-EVAL framework assesses LLM deployment viability via energy, latency, cost constraints beyond accuracy metrics.

Framework for scaling evaluation-driven discovery loops with LLMs using verifiers and simulators for hypothesis refinement.

LightGBM post-processing improves weather forecasts for precipitation, temperature, wind via feature selection on JMA data.

FedSEA framework extends federated online learning with stochastically extended adversary for privacy-preserving decentralized learning.

Empirical study shows active learning limitations for chemical reaction extraction despite uncertainty/diversity strategies.

Empty post with no content.

Framework evaluates LLM-generated parliamentary debate summaries using computational argumentation for faithfulness assessment.

Critique that agent/agentic tools like OpenClaw reduce utility for experienced developers despite onboarding novices.

Yelp is giving its chatbot assistant a major upgrade, turning the platform into something closer to a digital concierge with a suite of new features designed for "getting things done." The move, one of several AI-focused updates in recent months, is part of a broader industry push to make AI more relevant and practically useful to consumers while turning huge troves of user-generated data into a competitive edge. In a press release, Yelp says the Yelp Assistant chatbot will be at "the center of the app experience," where it can answer questions, make recommendations, and even handle bookings ...

Claude Auto mode in Code allows tool execution without per-call approval but lacks safety guardrails, enabling dangerous Bash commands.

PLaMo 2.1-VL: lightweight 2B/8B VLM for edge deployment with Japanese support, visual grounding, factory/infrastructure applications.

Rough set analysis reveals concept-level inconsistencies in Derm7pt dermoscopy dataset limit concept bottleneck model accuracy.

RDP LoRA uses geometry-driven trajectory analysis to identify optimal LoRA adapter placement in LLM fine-tuning without training.

Conditioning consistency gap quantifies KL divergence between Conditional Neural Process predictions when context points are added.

Co-Refine provides real-time AI feedback on coding drift in qualitative research without disrupting analyst workflow.

LLMs exhibit normative conformity in multi-agent discussions distinct from informational conformity, mechanistically analyzed via new tasks.

HalluAudio: first large-scale benchmark for hallucination detection across speech, environmental sound, and music in audio-language models.

Small Language Models (<10B params) deployed via agent paradigms (tool use, multi-agent) match larger models with reduced compute.

IndiaFinBench: first public benchmark for LLM evaluation on Indian financial regulatory text (SEBI, RBI documents).

DOPE: semiparametric estimator for scalar functionals from neural operator solution trajectories with debiasing.

TEMPO scales test-time training for LLMs via periodic critic recalibration to avoid reward drift and diversity collapse.

Tim Cook's departure as Apple CEO analyzed through lens of timing and business cycles; not AI-specific.

LocQA benchmark (2,156 locale-ambiguous questions in 12 languages) exposes inter/intra-lingual biases in multilingual LLMs on facts, laws, and measurements.

VB-Score evaluation framework for medical QA systems assesses accuracy and health equity risks beyond semantic similarity matching.

Explicit Trait Inference (ETI) method uses warmth/competence dimensions to improve LLM-based multi-agent coordination, reducing payoff loss by 45-77%.

HarDBench benchmark evaluates LLM robustness against draft-based co-authoring jailbreak attacks in collaborative writing scenarios.

Analysis of OpenRouter token usage shows top coding apps predominantly non-coding models; Hermes adoption noted.

Contradictory user reports on Claude Opus 4.7: some praise instruction adherence and comprehension; others cite regression in context and creativity.

CulturALL benchmark assesses LLMs on grounded multilingual/multicultural tasks via human-AI collaborative framework across real-world scenarios.