https://aiexpert.news/en/article/evaluating-agents-under-real-constraints-the-grounded-reasoning-cup ai|expert en 2026-08-18T10:50:37.366Z Stanford Wins Reasoning Cup With System Design, Not Model Swap https://aiexpert.news/pt/article/evaluating-agents-under-real-constraints-the-grounded-reasoning-cup ai|expert pt-BR 2026-08-18T10:50:37.366Z Stanford Vence a Reasoning Cup Com Design de Sistema, Não Troca de Modelo https://aiexpert.news/es/article/evaluating-agents-under-real-constraints-the-grounded-reasoning-cup ai|expert es 2026-08-18T10:50:37.366Z Stanford Gana la Reasoning Cup Con Diseño de Sistema, No Intercambio de Modelos https://aiexpert.news/en/article/baton-cost-efficient-agent-architectures-via-selective-model-invocation ai|expert en 2026-08-18T07:46:37.612Z BATON Cuts Robot Task Exploration Cost from Exponential to Linear https://aiexpert.news/pt/article/baton-cost-efficient-agent-architectures-via-selective-model-invocation ai|expert pt-BR 2026-08-18T07:46:37.612Z BATON reduz o custo de exploração de tarefas de robôs de exponencial para linear https://aiexpert.news/es/article/baton-cost-efficient-agent-architectures-via-selective-model-invocation ai|expert es 2026-08-18T07:46:37.612Z BATON reduce el costo de exploración de tareas de robot de exponencial a lineal https://aiexpert.news/en/article/anthropic-and-openai-collaborate-on-agent-tool-standardization-ports-pike ai|expert en 2026-08-18T07:18:37.311Z OpenAI and NVIDIA Lock 8-Gigawatt AI Factory to Ohio Campus https://aiexpert.news/pt/article/anthropic-and-openai-collaborate-on-agent-tool-standardization-ports-pike ai|expert pt-BR 2026-08-18T07:18:37.311Z OpenAI e NVIDIA Garantem Fábrica de IA de 8 Gigawatts no Campus de Ohio https://aiexpert.news/es/article/anthropic-and-openai-collaborate-on-agent-tool-standardization-ports-pike ai|expert es 2026-08-18T07:18:37.311Z OpenAI y NVIDIA Aseguran Fábrica de IA de 8 Gigawatts en Campus de Ohio https://aiexpert.news/en/article/proteus-incremental-memory-activation-cuts-long-context-cost ai|expert en 2026-08-18T06:38:37.748Z Memory Allocation Fix Lets Recurrent Models Match Attention at Scale https://aiexpert.news/pt/article/proteus-incremental-memory-activation-cuts-long-context-cost ai|expert pt-BR 2026-08-18T06:38:37.748Z Correção de Alocação de Memória Permite que Modelos Recorrentes Correspondam à Atenção em Escala https://aiexpert.news/es/article/proteus-incremental-memory-activation-cuts-long-context-cost ai|expert es 2026-08-18T06:38:37.748Z Corrección de Asignación de Memoria Permite que Modelos Recurrentes Igualen Atención a Escala https://aiexpert.news/en/article/ab-testing-models-in-production-safe-canary-patterns-for-model-rollouts ai|expert en 2026-08-18T04:52:37.607Z A/B testing models moves from app code to serving layer https://aiexpert.news/pt/article/ab-testing-models-in-production-safe-canary-patterns-for-model-rollouts ai|expert pt-BR 2026-08-18T04:52:37.607Z Testes A/B de modelos saem do código da aplicação para a camada de serving https://aiexpert.news/es/article/ab-testing-models-in-production-safe-canary-patterns-for-model-rollouts ai|expert es 2026-08-18T04:52:37.607Z Las pruebas A/B de modelos se trasladan del código de la aplicación a la capa de serving https://aiexpert.news/en/article/model-hypnosis-frontier-models-hijackable-via-subliminal-prompt-cues ai|expert en 2026-08-18T04:20:37.397Z Researchers Expose Model Hypnosis: Invisible Prompt Cues Hijack AI With 99% Success https://aiexpert.news/pt/article/model-hypnosis-frontier-models-hijackable-via-subliminal-prompt-cues ai|expert pt-BR 2026-08-18T04:20:37.397Z Pesquisadores Expõem Hipnose de Modelo: Dicas de Prompt Invisíveis Sequestram IA com 99% de Sucesso https://aiexpert.news/es/article/model-hypnosis-frontier-models-hijackable-via-subliminal-prompt-cues ai|expert es 2026-08-18T04:20:37.397Z Investigadores Exponen Hipnosis de Modelo: Señales de Prompt Invisibles Secuestran IA con un 99% de Éxito https://aiexpert.news/en/article/deployed-compliance-detectors-are-rule-blind-audit-finds ai|expert en 2026-08-18T03:30:23.909Z Compliance Detectors Fail Rule-Blindness Test, Lexsi Audit Finds https://aiexpert.news/pt/article/deployed-compliance-detectors-are-rule-blind-audit-finds ai|expert pt-BR 2026-08-18T03:30:23.909Z Detectores de Conformidade Falham no Teste de Cegueira de Regras, Auditoria Lexsi Constata https://aiexpert.news/es/article/deployed-compliance-detectors-are-rule-blind-audit-finds ai|expert es 2026-08-18T03:30:23.909Z Detectores de Cumplimiento Fallan en la Prueba de Ceguera de Reglas, Auditoría Lexsi Encuentra https://aiexpert.news/en/article/production-serving-stacks-need-better-risk-budgets-for-quality-latency-trade-off ai|expert en 2026-08-18T02:58:38.079Z Wei-Liu Paper Halves Production Serving Fallback Rate With Anytime-Valid Ledgers https://aiexpert.news/pt/article/production-serving-stacks-need-better-risk-budgets-for-quality-latency-trade-off ai|expert pt-BR 2026-08-18T02:58:38.079Z Artigo de Wei-Liu Reduz à Metade a Taxa de Fallback em Production Serving Com Ledgers Anytime-Valid https://aiexpert.news/es/article/production-serving-stacks-need-better-risk-budgets-for-quality-latency-trade-off ai|expert es 2026-08-18T02:58:38.079Z Paper de Wei-Liu Reduce a la Mitad la Tasa de Fallback en Production Serving Con Ledgers Anytime-Valid https://aiexpert.news/en/article/how-a-7b-model-matches-13b-performance-on-long-context-reasoning-tasks ai|expert en 2026-08-18T02:26:37.611Z KV-Rescue Recovers 87% of Lost Model Accuracy https://aiexpert.news/pt/article/how-a-7b-model-matches-13b-performance-on-long-context-reasoning-tasks ai|expert pt-BR 2026-08-18T02:26:37.611Z KV-Rescue Recupera 87% da Precisão Perdida do Modelo https://aiexpert.news/es/article/how-a-7b-model-matches-13b-performance-on-long-context-reasoning-tasks ai|expert es 2026-08-18T02:26:37.611Z KV-Rescue Recupera el 87% de la Precisión Perdida del Modelo https://aiexpert.news/en/article/serving-ml-features-in-sub-second-latency-databricks-new-production-freshness-ta ai|expert en 2026-08-17T23:16:37.551Z Databricks Feature Store Cuts ML Feature Latency to 200ms https://aiexpert.news/pt/article/serving-ml-features-in-sub-second-latency-databricks-new-production-freshness-ta ai|expert pt-BR 2026-08-17T23:16:37.551Z Databricks Feature Store Reduz Latência de Features de ML para 200ms https://aiexpert.news/es/article/serving-ml-features-in-sub-second-latency-databricks-new-production-freshness-ta ai|expert es 2026-08-17T23:16:37.551Z Databricks Feature Store Reduce la Latencia de Features de ML a 200ms https://aiexpert.news/en/article/openai-on-the-defenders-window-robustness-vs-deployability-trade-off ai|expert en 2026-08-17T21:20:37.232Z OpenAI Races to Weaponize Defenses Before AI Goes Open https://aiexpert.news/pt/article/openai-on-the-defenders-window-robustness-vs-deployability-trade-off ai|expert pt-BR 2026-08-17T21:20:37.232Z OpenAI se Corre para Armadilhar Defesas Antes da IA Ficar Aberta https://aiexpert.news/es/article/openai-on-the-defenders-window-robustness-vs-deployability-trade-off ai|expert es 2026-08-17T21:20:37.232Z OpenAI se Apresura a Armar Defensas Antes de que la IA Sea Abierta https://aiexpert.news/en/article/gpu-clusters-gain-33-points-utilization-via-task-ordering-without-new-hardware ai|expert en 2026-08-17T20:48:37.357Z Reordered GPU Scheduling Lifts Utilization 33 Percentage Points https://aiexpert.news/pt/article/gpu-clusters-gain-33-points-utilization-via-task-ordering-without-new-hardware ai|expert pt-BR 2026-08-17T20:48:37.357Z Reordenação de Scheduling de GPU Eleva Utilização em 33 Pontos Percentuais https://aiexpert.news/es/article/gpu-clusters-gain-33-points-utilization-via-task-ordering-without-new-hardware ai|expert es 2026-08-17T20:48:37.357Z La Reordenación de Programación de GPU Eleva la Utilización 33 Puntos Porcentuales https://aiexpert.news/en/article/grab-cuts-mechanical-analytics-work-from-44-to-30-with-ai-agent-delegation ai|expert en 2026-08-17T20:16:38.399Z Grab cuts mechanical analytics work to 30% with agent delegation https://aiexpert.news/pt/article/grab-cuts-mechanical-analytics-work-from-44-to-30-with-ai-agent-delegation ai|expert pt-BR 2026-08-17T20:16:38.399Z Grab reduz trabalho em análise mecânica para 30% com delegação de agentes https://aiexpert.news/es/article/grab-cuts-mechanical-analytics-work-from-44-to-30-with-ai-agent-delegation ai|expert es 2026-08-17T20:16:38.399Z Grab reduce trabajo en análisis mecánico al 30% con delegación de agentes