ニュースハブ

GLiGuard uses schema-conditioned classification to moderate LLM-generated content.

GLiNER2-PII extracts personally identifiable information across multiple languages.

Method corrects stochastic update bias in preconditioned language model optimizers.

Framework measures proactive problem-solving capabilities in LLM agents beyond reactive behavior.

GLiNER2 performs multi-task information extraction using schema-driven interface design.

GLiNER is a generalist named entity recognition model using bidirectional transformers.

Pioneer Agent continuously improves small language models deployed in production.

LangChain offers managed deep agents with durable execution and observability in private beta.

LangSmith LLM Gateway provides runtime governance with spend limits and PII redaction.

Managed Deep Agents now publicly available with durable execution and production infrastructure.

OpenAI released preliminary cybersecurity evaluations for Astra and described safeguards and security control improvements.

HSP GRUPPE uses ChatGPT Enterprise to increase productivity and capacity in tax advisory services.

TutorMoments evaluates whether AI tutors recognize when to support students versus encourage independent reasoning.

Deep Agents, LangChain, and LangGraph offer distinct approaches for building agents with different use cases.

Anthropic improved safety measures in Claude 5 to prevent misuse of biological information.

Study compares performance of diffusion-based and autoregressive language models across NLP tasks.

Arbitrage uses advantage-aware speculation to reduce computational cost of long chain-of-thought reasoning in LLMs.

Decode Context Parallelism in vLLM achieves 3x higher throughput on long-context agentic workloads.

Categorical flow maps enable continuous diffusion models for discrete data generation with faster sampling.

Tencent Hunyuan contributes high-performance attention, router GEMM, and MoE kernels to SGLang.