News-Hub

Google DeepMind introduced agentic video understanding capabilities for Gemini.

NVIDIA Nemotron enables adaptive agentic systems for cybersecurity operations.

Basis, Clay, and Exa Labs use AI agents to automate enterprise workflows.

NVIDIA provides guidance on sizing GPUs for AI inference and cost optimization.

ChatGPT can now securely connect to healthcare EHR and medical research data.

vLLM-Omni optimizes MiniMax H3 and integrates FastVideo's FastH3 for video generation faster than real-time playback.

Hugging Face released 200+ WebGPU kernels for running AI models locally.

Context-Aware Interleaved Batching maintains historical context during batched speech transcription to improve punctuation and terminology accuracy.

Configurable semantic chunking for biomedical RAG combines entity-preserving windows and trigger-centered chunking to avoid fragmenting semantic evidence.

OntoAligner-Ensemble uses voting-based fusion to reconcile outputs from heterogeneous ontology alignment techniques including LLMs and embeddings.

DIASENTINEL is a multi-agent system for diabetes risk screening from electronic health records with guideline-grounded report generation.

Controlled evaluation of 13 LLMs across Qwen and GPT variants shows varying effects of model scale on ontology learning performance.

Aspire enables LLMs to self-evolve from vague goals by interpreting objectives, identifying capability gaps, and assessing improvement.

BLOOM-WILT uses logit tilting to improve sample efficiency of automated LLM auditors in detecting deployed model behaviors.

Industrial LLM post-training treats deployed checkpoints as dataware artifacts updated via bounded mixture patches under fixed compute budgets.

S3Gym evaluates whether LLMs can self-test, self-judge, and self-improve through behavioral experience accumulated in environments.

J-lens verbalization uses first tokens as clues to represent multi-token concepts without requiring precomputed phrase vocabularies.

Method reduces token consumption for LLM agents reasoning over unstructured data by adaptively structuring documents.

Technique combines process supervision with outcome-based credit assignment for improved agent policy optimization.

System automatically induces evaluation rubrics to guide autonomous research agents on open-ended tasks.