https://aiexpert.news/en/article/steering-reasoning-models-by-predicting-future-misbehavior ai|expert en 2026-06-10T15:20:10.039Z FPCG steers reasoning models at test time without retraining https://aiexpert.news/pt/article/steering-reasoning-models-by-predicting-future-misbehavior ai|expert pt-BR 2026-06-10T15:20:10.039Z FPCG dirige modelos de raciocínio em tempo de teste sem retreinamento https://aiexpert.news/es/article/steering-reasoning-models-by-predicting-future-misbehavior ai|expert es 2026-06-10T15:20:10.039Z FPCG dirige modelos de razonamiento en tiempo de prueba sin reentrenamiento https://aiexpert.news/en/article/predictive-signals-improve-steering-of-reasoning-models ai|expert en 2026-06-10T14:40:10.592Z Linear Probes Predict Reasoning-Model Behavior at 64–91% Accuracy https://aiexpert.news/pt/article/predictive-signals-improve-steering-of-reasoning-models ai|expert pt-BR 2026-06-10T14:40:10.592Z Sondas Lineares Atingem 64-91% de Precisão para Modelos de Raciocínio Direcionados https://aiexpert.news/es/article/predictive-signals-improve-steering-of-reasoning-models ai|expert es 2026-06-10T14:40:10.592Z Sondeos Lineales Logran Precisión del 64-91% en Modelos de Razonamiento https://aiexpert.news/en/article/llm-benchmarks-fail-to-capture-reliability-and-failure-modes ai|expert en 2026-06-10T14:02:10.363Z LLM Leaderboards Fail to Predict Production Reliability https://aiexpert.news/pt/article/llm-benchmarks-fail-to-capture-reliability-and-failure-modes ai|expert pt-BR 2026-06-10T14:02:10.363Z Quadros de Liderança de LLM Não Prevêem Confiabilidade em Produção https://aiexpert.news/es/article/llm-benchmarks-fail-to-capture-reliability-and-failure-modes ai|expert es 2026-06-10T14:02:10.363Z Las clasificaciones de LLM no predicen fiabilidad en producción https://aiexpert.news/en/article/databricks-enhances-amlbsa-compliance-for-data-platforms ai|expert en 2026-06-10T13:28:10.062Z Databricks Introduces 8-10x Faster AML Processing, But Lacks Customer Proof https://aiexpert.news/pt/article/databricks-enhances-amlbsa-compliance-for-data-platforms ai|expert pt-BR 2026-06-10T13:28:10.062Z Databricks Introduce 8-10x Mais Rápido Processamento AML, Mas Sem Prova de Clientes https://aiexpert.news/es/article/databricks-enhances-amlbsa-compliance-for-data-platforms ai|expert es 2026-06-10T13:28:10.062Z Databricks Introduce 8-10 veces más rápido en Procesamiento AML, pero sin Prueba de Clientes https://aiexpert.news/en/article/abc-bench-measuring-biosecurity-risks-of-autonomous-ai-agents-in-research ai|expert en 2026-06-10T12:20:10.431Z Grok 3 Surpasses Credentialed Biologists on Autonomous DNA Lab Tasks https://aiexpert.news/pt/article/abc-bench-measuring-biosecurity-risks-of-autonomous-ai-agents-in-research ai|expert pt-BR 2026-06-10T12:20:10.431Z Grok 3 Supera Biologistas Credenciados em Tarefas de Laboratório Automático de DNA https://aiexpert.news/es/article/abc-bench-measuring-biosecurity-risks-of-autonomous-ai-agents-in-research ai|expert es 2026-06-10T12:20:10.431Z Grok 3 Supera a Biólogos Credenciados en Tareas de Laboratorio ADN Autónomo https://aiexpert.news/en/article/azure-apim-introduces-unified-model-routing-and-mcp-content-safety ai|expert en 2026-06-10T11:38:10.356Z Azure Extends Content Safety to Agent-to-Agent AI Traffic https://aiexpert.news/pt/article/azure-apim-introduces-unified-model-routing-and-mcp-content-safety ai|expert pt-BR 2026-06-10T11:38:10.356Z Azure Amplia Segurança de Conteúdo para Tráfego de Agente-a-Agente IA https://aiexpert.news/es/article/azure-apim-introduces-unified-model-routing-and-mcp-content-safety ai|expert es 2026-06-10T11:38:10.356Z Azure Amplía la Seguridad de Contenido al Tráfico de Agente a Agente de IA https://aiexpert.news/en/article/azure-api-management-ships-unified-model-api-layer-and-mcp ai|expert en 2026-06-10T11:00:18.254Z Azure's Unified Model API Now Routes Requests to Any LLM Without Client Rewrites https://aiexpert.news/pt/article/azure-api-management-ships-unified-model-api-layer-and-mcp ai|expert pt-BR 2026-06-10T11:00:18.254Z API de Modelo Unificado do Azure Redireciona Solicitações para Qualquer LLM Sem Reescrever o Cliente https://aiexpert.news/es/article/azure-api-management-ships-unified-model-api-layer-and-mcp ai|expert es 2026-06-10T11:00:18.254Z La API de Modelo Unificada de Azure Enruta Solicitudes a Cualquier LLM Sin Reescribir el Cliente https://aiexpert.news/en/article/microsofts-unified-model-api-gateway-integrates-mcp-for-safety ai|expert en 2026-06-10T10:20:10.287Z Microsoft's API Gateway Unifies Model Access Across Five Providers https://aiexpert.news/pt/article/microsofts-unified-model-api-gateway-integrates-mcp-for-safety ai|expert pt-BR 2026-06-10T10:20:10.287Z Gateway de API da Microsoft Unifica Acesso a Modelos entre Cinco Provedores https://aiexpert.news/es/article/microsofts-unified-model-api-gateway-integrates-mcp-for-safety ai|expert es 2026-06-10T10:20:10.287Z Puerta de API de Microsoft Unifica el Acceso a Modelos entre Cinco Proveedores https://aiexpert.news/en/article/eevee-self-improving-agents-that-adapt-prompts-in-real-time-across-domains ai|expert en 2026-06-10T09:20:10.194Z EEVEE Surpasses Self-Improving Agents with 48% Margin on Multi-Domain Inference https://aiexpert.news/pt/article/eevee-self-improving-agents-that-adapt-prompts-in-real-time-across-domains ai|expert pt-BR 2026-06-10T09:20:10.194Z EEVEE Vence Agentes de Auto-Melhoria com Margem de 48% em Inferência Multi-Domínio https://aiexpert.news/es/article/eevee-self-improving-agents-that-adapt-prompts-in-real-time-across-domains ai|expert es 2026-06-10T09:20:10.194Z EEVEE supera agentes de auto-mejora con margen del 48% en inferencia multi-dominio https://aiexpert.news/en/article/cohere-launches-north-mini-code-smaller-open-weights-code-model ai|expert en 2026-06-10T07:46:10.258Z Cohere's North Mini Code Tops Rivals on Single GPU https://aiexpert.news/pt/article/cohere-launches-north-mini-code-smaller-open-weights-code-model ai|expert pt-BR 2026-06-10T07:46:10.258Z Cohere's North Mini Code Supera Rivais em Única GPU https://aiexpert.news/es/article/cohere-launches-north-mini-code-smaller-open-weights-code-model ai|expert es 2026-06-10T07:46:10.258Z Cohere's North Mini Code supera a competidores en una sola GPU https://aiexpert.news/en/article/servicenows-code-switching-benchmark-stress-testing-asr-for-multilingual-agents ai|expert en 2026-06-10T07:14:10.130Z ElevenLabs Scribe Tops Code-Switched Speech Benchmark https://aiexpert.news/pt/article/servicenows-code-switching-benchmark-stress-testing-asr-for-multilingual-agents ai|expert pt-BR 2026-06-10T07:14:10.130Z ElevenLabs Scribe Supera Benchmark de Fala com Troca de Código https://aiexpert.news/es/article/servicenows-code-switching-benchmark-stress-testing-asr-for-multilingual-agents ai|expert es 2026-06-10T07:14:10.130Z ElevenLabs Scribe lidera el marco de referencia de voz intercambiada en código https://aiexpert.news/en/article/piper-programmable-system-for-composable-parallelism-in-model-training ai|expert en 2026-06-10T05:50:10.095Z University of Washington's Piper compiler unifies distributed training schedules https://aiexpert.news/pt/article/piper-programmable-system-for-composable-parallelism-in-model-training ai|expert pt-BR 2026-06-10T05:50:10.095Z Compilador Piper da Universidade de Washington unifica agendas de treinamento distribuído https://aiexpert.news/es/article/piper-programmable-system-for-composable-parallelism-in-model-training ai|expert es 2026-06-10T05:50:10.095Z El compilador Piper de la Universidad de Washington unifica los horarios de entrenamiento distribuido https://aiexpert.news/en/article/piper-programmable-distributed-training-removes-manual-strategy-design ai|expert en 2026-06-10T05:16:10.040Z Piper compiler enables DeepSeek-style training at thousand-GPU scale https://aiexpert.news/pt/article/piper-programmable-distributed-training-removes-manual-strategy-design ai|expert pt-BR 2026-06-10T05:16:10.040Z Compilador Piper permite treinamento estilo DeepSeek em escala de milhares de GPUs https://aiexpert.news/es/article/piper-programmable-distributed-training-removes-manual-strategy-design ai|expert es 2026-06-10T05:16:10.040Z El compilador Piper permite el entrenamiento estilo DeepSeek a escala de miles de GPU https://aiexpert.news/en/article/entergy-ceo-disputes-ai-data-center-electricity-bill-hikes ai|expert en 2026-06-10T04:32:10.417Z $7 Billion Savings Mask Entergy Data Center Operators' Long-Term Infrastructure Debt https://aiexpert.news/pt/article/entergy-ceo-disputes-ai-data-center-electricity-bill-hikes ai|expert pt-BR 2026-06-10T04:32:10.417Z Poupança de US$ 7 Bilhões Oculta Dívida de Infraestrutura a Longo Prazo dos Operadores do Centro de Dados Entergy https://aiexpert.news/es/article/entergy-ceo-disputes-ai-data-center-electricity-bill-hikes ai|expert es 2026-06-10T04:32:10.417Z Ahorro de $7 mil millones enmascaran la deuda a largo plazo de infraestructura de los operadores de centros de datos de Entergy https://aiexpert.news/en/article/piper-a-programmable-system-for-distributed-ml-training ai|expert en 2026-06-10T03:42:10.199Z Piper Compiler Eliminates Hand-Coding for Distributed Training https://aiexpert.news/pt/article/piper-a-programmable-system-for-distributed-ml-training ai|expert pt-BR 2026-06-10T03:42:10.199Z Compilador Piper Elimina a Codificação Manual para Treinamento Distribuído https://aiexpert.news/es/article/piper-a-programmable-system-for-distributed-ml-training ai|expert es 2026-06-10T03:42:10.199Z Compilador Piper Elimina la Codificación Manual para Entrenamiento Distribuido https://aiexpert.news/en/article/multi-token-inference-has-a-hidden-architectural-flawclp-proposes-a-fix ai|expert en 2026-06-10T02:10:11.306Z Single Linear Layer Outperforms 1M-Parameter Gate in MTP Speedup Test https://aiexpert.news/pt/article/multi-token-inference-has-a-hidden-architectural-flawclp-proposes-a-fix ai|expert pt-BR 2026-06-10T02:10:11.306Z Camada Linear Única Supera Porta de 1M-Parâmetros no Teste de Aceleração de MTP https://aiexpert.news/es/article/multi-token-inference-has-a-hidden-architectural-flawclp-proposes-a-fix ai|expert es 2026-06-10T02:10:11.306Z Capa Lineal Simple Supera Puerta de 1M-Parámetros en Prueba de Aceleración MTP https://aiexpert.news/en/article/clinenv-real-ehr-data-benchmark-for-agent-decision-making-under-uncertainty ai|expert en 2026-06-10T01:00:19.054Z Real EHR Benchmark Exposes Limits of LLMs in Clinical Action https://aiexpert.news/pt/article/clinenv-real-ehr-data-benchmark-for-agent-decision-making-under-uncertainty ai|expert pt-BR 2026-06-10T01:00:19.054Z Real EHR Benchmark Exposta Limites dos LLMs em Ações Clínicas https://aiexpert.news/es/article/clinenv-real-ehr-data-benchmark-for-agent-decision-making-under-uncertainty ai|expert es 2026-06-10T01:00:19.054Z EHC Real Benchmark Revela Límites de LLM en Acción Clínica https://aiexpert.news/en/article/microsoft-backed-d-matrix-chips-challenge-nvidia-in-ai-inference ai|expert en 2026-06-09T13:50:10.288Z Corsair inference accelerator cuts response time 12× in GPU hybrid setup https://aiexpert.news/pt/article/microsoft-backed-d-matrix-chips-challenge-nvidia-in-ai-inference ai|expert pt-BR 2026-06-09T13:50:10.288Z Aceleração de inferência Corsair reduz tempo de resposta 12× em configuração híbrida de GPU https://aiexpert.news/es/article/microsoft-backed-d-matrix-chips-challenge-nvidia-in-ai-inference ai|expert es 2026-06-09T13:50:10.288Z Acelerador de inferencia Corsair reduce el tiempo de respuesta 12× en configuración híbrida de GPU https://aiexpert.news/en/article/aha-wam-decoupling-temporal-resolutions-in-world-action-models ai|expert en 2026-06-09T12:14:10.330Z AHA-WAM achieves 4.59× faster robot control by decoupling Diffusion Transformers https://aiexpert.news/pt/article/aha-wam-decoupling-temporal-resolutions-in-world-action-models ai|expert pt-BR 2026-06-09T12:14:10.330Z AHA-WAM alcança controle de robô 4.59 vezes mais rápido desacoplando Transformadores de Difusão https://aiexpert.news/es/article/aha-wam-decoupling-temporal-resolutions-in-world-action-models ai|expert es 2026-06-09T12:14:10.330Z AHA-WAM logra un control de robot 4.59 veces más rápido al desacoplar Transformadores de Difusión https://aiexpert.news/en/article/fase-detecting-hallucinations-in-multi-agent-code-generation-at-scale ai|expert en 2026-06-09T10:50:10.398Z FASE Cuts Hallucination Detection to 333x Speed https://aiexpert.news/pt/article/fase-detecting-hallucinations-in-multi-agent-code-generation-at-scale ai|expert pt-BR 2026-06-09T10:50:10.398Z FASE Reduz Detecção de Alucinações para 333x Velocidade https://aiexpert.news/es/article/fase-detecting-hallucinations-in-multi-agent-code-generation-at-scale ai|expert es 2026-06-09T10:50:10.398Z FASE Reduce la Detección de Alucinaciones a una Velocidad de 333x https://aiexpert.news/en/article/llm-rl-trust-region-fix-tackles-ppos-long-tail-failure-mode ai|expert en 2026-06-09T09:28:10.100Z New DRPO Method Fixes Long-Tail Vocabulary Collapse in LLM RL https://aiexpert.news/pt/article/llm-rl-trust-region-fix-tackles-ppos-long-tail-failure-mode ai|expert pt-BR 2026-06-09T09:28:10.100Z Novo Método DRPO Corrige Colapso de Vocabulário de Longo-Cabeça em RL de LLM https://aiexpert.news/es/article/llm-rl-trust-region-fix-tackles-ppos-long-tail-failure-mode ai|expert es 2026-06-09T09:28:10.100Z Nuevo Método DRPO Corrige Colapso de Vocabulario de Larga-Talla en RL de LLM https://aiexpert.news/en/article/arxiv-fase-reduces-llm-hallucination-uncertainty-in-multi-agent-code-gen-without ai|expert en 2026-06-09T08:56:09.989Z FASE Cuts Hallucination Detection Cost to 0.3% of Rivals https://aiexpert.news/pt/article/arxiv-fase-reduces-llm-hallucination-uncertainty-in-multi-agent-code-gen-without ai|expert pt-BR 2026-06-09T08:56:09.989Z FASE Reduz Custo de Detecção de Alucinações para 0,3% dos Concorrentes https://aiexpert.news/es/article/arxiv-fase-reduces-llm-hallucination-uncertainty-in-multi-agent-code-gen-without ai|expert es 2026-06-09T08:56:09.989Z FASE Reduce el Costo de Detección de Alucinaciones al 0.3% de los Rivales https://aiexpert.news/en/article/siga-adapting-coding-agents-to-scientific-simulators-without-domain-fine-tuning ai|expert en 2026-06-09T08:24:10.237Z SIGA Speeds Coding Agents on Scientific Simulators by 36× https://aiexpert.news/pt/article/siga-adapting-coding-agents-to-scientific-simulators-without-domain-fine-tuning ai|expert pt-BR 2026-06-09T08:24:10.237Z SIGA Acelera Agentes de Codificação em Simuladores Científicos por 36× https://aiexpert.news/es/article/siga-adapting-coding-agents-to-scientific-simulators-without-domain-fine-tuning ai|expert es 2026-06-09T08:24:10.237Z SIGA Acelera Agentes de Codificación en Simuladores Científicos por 36× https://aiexpert.news/en/article/echo-memory-isolating-object-persistence-failures-in-world-models ai|expert en 2026-06-09T07:38:10.210Z Echo-Memory Shows World Models Fail the Revisit Test https://aiexpert.news/pt/article/echo-memory-isolating-object-persistence-failures-in-world-models ai|expert pt-BR 2026-06-09T07:38:10.210Z Echo-Memory Mostra que Modelos de Mundo Falham no Teste de Revisão https://aiexpert.news/es/article/echo-memory-isolating-object-persistence-failures-in-world-models ai|expert es 2026-06-09T07:38:10.210Z Echo-Memory Demuestra que los Modelos del Mundo Fallan en la Prueba de Revisita https://aiexpert.news/en/article/fase-fast-adaptive-semantic-entropy-for-multi-agent-code-reliability ai|expert en 2026-06-09T07:00:10.319Z Waterloo researchers cut uncertainty quantification cost 99.7% with FASE https://aiexpert.news/pt/article/fase-fast-adaptive-semantic-entropy-for-multi-agent-code-reliability ai|expert pt-BR 2026-06-09T07:00:10.319Z Pesquisadores de Waterloo reduzem custo de quantificação de incerteza em 99,7% com FASE https://aiexpert.news/es/article/fase-fast-adaptive-semantic-entropy-for-multi-agent-code-reliability ai|expert es 2026-06-09T07:00:10.319Z Investigadores de Waterloo reducen el costo de cuantificación de incertidumbre un 99.7% con FASE https://aiexpert.news/en/article/uk-sovereign-ai-push-with-nvidia-policy-meets-infrastructure ai|expert en 2026-06-09T06:24:10.003Z UK Deploys 120,000 Blackwell GPUs for Sovereign AI Compute https://aiexpert.news/pt/article/uk-sovereign-ai-push-with-nvidia-policy-meets-infrastructure ai|expert pt-BR 2026-06-09T06:24:10.003Z Reino Unido Utiliza 120.000 GPUs Blackwell para Computação de IA Soberana https://aiexpert.news/es/article/uk-sovereign-ai-push-with-nvidia-policy-meets-infrastructure ai|expert es 2026-06-09T06:24:10.003Z Reino Unido despliega 120,000 GPUs Blackwell para cómputo de IA soberano https://aiexpert.news/en/article/evaluation-cards-45-author-consortium-standardizes-ai-model-performance-reportin ai|expert en 2026-06-09T04:36:10.216Z EvalCards Schema Exposes Systematic AI Benchmark Metadata Gaps https://aiexpert.news/pt/article/evaluation-cards-45-author-consortium-standardizes-ai-model-performance-reportin ai|expert pt-BR 2026-06-09T04:36:10.216Z Esquema EvalCards Expõe Falhas Sistematizadas em Metadados de Benchmarks de IA https://aiexpert.news/es/article/evaluation-cards-45-author-consortium-standardizes-ai-model-performance-reportin ai|expert es 2026-06-09T04:36:10.216Z El Esquema EvalCards Revela Brechas Metadatos de Comparación de IA Sistematizadas https://aiexpert.news/en/article/perplexity-study-how-agentic-ai-reshapes-knowledge-work ai|expert en 2026-06-09T04:04:10.136Z Perplexity Agentic AI Cuts Task Time 87 Percent in Production Study https://aiexpert.news/pt/article/perplexity-study-how-agentic-ai-reshapes-knowledge-work ai|expert pt-BR 2026-06-09T04:04:10.136Z Perplexidade IA Agênica Reduz Tempo de Tarefa 87% em Estudo de Produção https://aiexpert.news/es/article/perplexity-study-how-agentic-ai-reshapes-knowledge-work ai|expert es 2026-06-09T04:04:10.136Z Agente de IA de Perplexity Reduce el Tiempo de Tareas un 87 Por Ciento en Estudio de Producción