https://aiexpert.news/en/article/selective-training-of-super-weights-fails-a-wake-up-call-for-model-compression ai|expert en 2026-07-10T23:24:23.383Z Super Weights Training Fails on OLMo Models, Demolishing Sparse Fine-Tuning Strategy https://aiexpert.news/pt/article/selective-training-of-super-weights-fails-a-wake-up-call-for-model-compression ai|expert pt-BR 2026-07-10T23:24:23.383Z Treinamento de Pesos Super falha em modelos OLMo, demolindo a estratégia de afinação esparsa https://aiexpert.news/es/article/selective-training-of-super-weights-fails-a-wake-up-call-for-model-compression ai|expert es 2026-07-10T23:24:23.383Z El entrenamiento de Super Weights falla en modelos OLMo, desmontando la estrategia de afinación esparsa https://aiexpert.news/en/article/smart-tiered-caching-across-cloud-regions-cloudflares-inference-latency-playbook ai|expert en 2026-07-10T22:52:23.666Z Cloudflare Cuts Latency Hundreds of Milliseconds With Smart Tiered Cache Update https://aiexpert.news/pt/article/smart-tiered-caching-across-cloud-regions-cloudflares-inference-latency-playbook ai|expert pt-BR 2026-07-10T22:52:23.666Z Cloudflare Reduz Latência em Centenas de Milissegundos com Atualização do Smart Tiered Cache https://aiexpert.news/es/article/smart-tiered-caching-across-cloud-regions-cloudflares-inference-latency-playbook ai|expert es 2026-07-10T22:52:23.666Z Cloudflare Reduce Latencia en Cientos de Milisegundos con Actualización de Smart Tiered Cache https://aiexpert.news/en/article/ai-industry-shifts-focus-from-scale-to-efficiency ai|expert en 2026-07-10T22:20:23.959Z 90% of Production Tokens Will Run on Open-Weight Models https://aiexpert.news/pt/article/ai-industry-shifts-focus-from-scale-to-efficiency ai|expert pt-BR 2026-07-10T22:20:23.959Z 90% dos Tokens de Produção Correrão em Modelos de Peso Aberto https://aiexpert.news/es/article/ai-industry-shifts-focus-from-scale-to-efficiency ai|expert es 2026-07-10T22:20:23.959Z El 90% de los Tokens de Producción Funcionará en Modelos de Peso Abierto https://aiexpert.news/en/article/us-eases-ai-export-controls-to-uae-new-mgx-partnership-signals-market-shift ai|expert en 2026-07-10T21:04:23.615Z U.S. Codifies the G42 Compliance Model as Standard for Allied AI Infrastructure https://aiexpert.news/pt/article/us-eases-ai-export-controls-to-uae-new-mgx-partnership-signals-market-shift ai|expert pt-BR 2026-07-10T21:04:23.615Z EUA Codificam o Modelo de Conformidade G42 como Padrão para Infraestrutura de IA Aliada https://aiexpert.news/es/article/us-eases-ai-export-controls-to-uae-new-mgx-partnership-signals-market-shift ai|expert es 2026-07-10T21:04:23.615Z EE. UU. Codifica el Modelo de Cumplimiento G42 como Estándar para la Infraestructura de IA Aliada https://aiexpert.news/en/article/databricks-details-data-layer-as-foundation-for-agentic-systems ai|expert en 2026-07-10T20:32:23.388Z Acxiom cuts pipeline run times 80-90% on Databricks agentic stack https://aiexpert.news/pt/article/databricks-details-data-layer-as-foundation-for-agentic-systems ai|expert pt-BR 2026-07-10T20:32:23.388Z Acxiom reduz tempo de execução de pipeline em 80-90% na pilha agentica do Databricks https://aiexpert.news/es/article/databricks-details-data-layer-as-foundation-for-agentic-systems ai|expert es 2026-07-10T20:32:23.388Z Acxiom reduce los tiempos de ejecución de la tubería entre un 80-90% en la pila agentica de Databricks https://aiexpert.news/en/article/slorr-stateless-low-rank-compression-for-efficient-neural-networks ai|expert en 2026-07-10T19:52:23.589Z Training-Efficient Low-Rank Compression Sidesteps Serving-Speed Proof https://aiexpert.news/pt/article/slorr-stateless-low-rank-compression-for-efficient-neural-networks ai|expert pt-BR 2026-07-10T19:52:23.589Z Compressão de Baixa Rank com Eficiência de Treinamento Desvia da Prova de Velocidade de Serviço https://aiexpert.news/es/article/slorr-stateless-low-rank-compression-for-efficient-neural-networks ai|expert es 2026-07-10T19:52:23.589Z Compresión de Rango Bajo Eficiente en Entrenamiento, Sin Pruebas de Velocidad en Servicio https://aiexpert.news/en/article/github-copilot-code-review-tools-improved-then-broke-performance ai|expert en 2026-07-10T19:20:23.535Z GitHub Copilot Reduces Code Review Costs 20% After Instruction Rewrite https://aiexpert.news/pt/article/github-copilot-code-review-tools-improved-then-broke-performance ai|expert pt-BR 2026-07-10T19:20:23.535Z O Copilot do GitHub Reduz Custos de Revisão de Código em 20% Após Reescrever Instruções https://aiexpert.news/es/article/github-copilot-code-review-tools-improved-then-broke-performance ai|expert es 2026-07-10T19:20:23.535Z GitHub Copilot Reduce el Costo de Revisión de Código un 20% Después de Reescribir Instrucciones https://aiexpert.news/en/article/chaos-engineering-for-gpu-clusters-hardening-production-inference-reliability ai|expert en 2026-07-10T18:00:52.393Z Fault-Injection Discipline Essential for GPU Production Reliability https://aiexpert.news/pt/article/chaos-engineering-for-gpu-clusters-hardening-production-inference-reliability ai|expert pt-BR 2026-07-10T18:00:52.393Z Disciplina de Injeção de Falhas Essencial para Confiabilidade em Produção de GPU https://aiexpert.news/es/article/chaos-engineering-for-gpu-clusters-hardening-production-inference-reliability ai|expert es 2026-07-10T18:00:52.393Z Disciplina de Inyección de Fallos Esencial para la Confiabilidad en Producción de GPU https://aiexpert.news/en/article/hugging-face-attention-profiling-guide-unlocks-2040-inference-throughput-gains ai|expert en 2026-07-10T17:30:23.381Z Hugging Face Cuts Inference Attention Overhead 20-40% With Fused Kernels https://aiexpert.news/pt/article/hugging-face-attention-profiling-guide-unlocks-2040-inference-throughput-gains ai|expert pt-BR 2026-07-10T17:30:23.381Z Hugging Face Reduz Sobrecarga de Atenção na Inferência em 20-40% com Núcleos Fundidos https://aiexpert.news/es/article/hugging-face-attention-profiling-guide-unlocks-2040-inference-throughput-gains ai|expert es 2026-07-10T17:30:23.381Z Hugging Face Reduce el Exceso de Atención en Inferencia 20-40% con Núcleos Fusionados https://aiexpert.news/en/article/sk-hynixs-265b-ipo-to-boost-hbm-production ai|expert en 2026-07-10T16:28:24.587Z Lock HBM Supply Now: SK Hynix IPO Buys Capacity, Not Time https://aiexpert.news/pt/article/sk-hynixs-265b-ipo-to-boost-hbm-production ai|expert pt-BR 2026-07-10T16:28:24.587Z Bloqueie o Fornecimento de HBM Agora: A IPO da SK Hynix Compra Capacidade, Não Tempo https://aiexpert.news/es/article/sk-hynixs-265b-ipo-to-boost-hbm-production ai|expert es 2026-07-10T16:28:24.587Z Bloquee el Suministro de HBM Ahora: La OPV de SK Hynix Compra Capacidad, No Tiempo https://aiexpert.news/en/article/slacks-agentic-e2e-ui-testing-harness-reduces-test-failures ai|expert en 2026-07-10T15:36:23.519Z Agentic Testing Reveals Cost Bottleneck, Not CI Solution https://aiexpert.news/pt/article/slacks-agentic-e2e-ui-testing-harness-reduces-test-failures ai|expert pt-BR 2026-07-10T15:36:23.519Z Testes Agentes revelam gargalo de custo, não solução de CI https://aiexpert.news/es/article/slacks-agentic-e2e-ui-testing-harness-reduces-test-failures ai|expert es 2026-07-10T15:36:23.519Z Pruebas Agenticas revelan cuello de botella de costo, no solución de CI https://aiexpert.news/en/article/uniclawbench-measuring-proactive-agent-reliability-at-scale ai|expert en 2026-07-10T14:34:23.685Z Claude Opus Fails Half of Real-World Tasks in UniClawBench https://aiexpert.news/pt/article/uniclawbench-measuring-proactive-agent-reliability-at-scale ai|expert pt-BR 2026-07-10T14:34:23.685Z Claude Opus Não Consegue Metade das Tarefas do Mundo Real no UniClawBench https://aiexpert.news/es/article/uniclawbench-measuring-proactive-agent-reliability-at-scale ai|expert es 2026-07-10T14:34:23.685Z Claude Opus Falla la Mitad de las Tarefas del Mundo Real en UniClawBench https://aiexpert.news/en/article/datadogs-test-driven-production-code-migration-with-claude-and-cursor ai|expert en 2026-07-10T14:02:23.679Z Datadog Cut Stream Router Runtime from 45 Minutes to One Second https://aiexpert.news/pt/article/datadogs-test-driven-production-code-migration-with-claude-and-cursor ai|expert pt-BR 2026-07-10T14:02:23.679Z Datadog reduz tempo de execução do Stream Router de 45 minutos para um segundo https://aiexpert.news/es/article/datadogs-test-driven-production-code-migration-with-claude-and-cursor ai|expert es 2026-07-10T14:02:23.679Z Datadog redujo el tiempo de ejecución de Stream Router de 45 minutos a un segundo https://aiexpert.news/en/article/githubs-solution-for-durable-repository-ownership ai|expert en 2026-07-09T19:50:23.686Z GitHub Ended 45-Day Repository Ownership Crisis With Queryable Properties https://aiexpert.news/pt/article/githubs-solution-for-durable-repository-ownership ai|expert pt-BR 2026-07-09T19:50:23.686Z GitHub Encerrou Crise de Propriedade de Repositório de 45 Dias com Propriedades Consultáveis https://aiexpert.news/es/article/githubs-solution-for-durable-repository-ownership ai|expert es 2026-07-09T19:50:23.686Z GitHub Resolvió Crisis de Propiedad de Repositorio de 45 Días con Propiedades Consultables https://aiexpert.news/en/article/grok-45-enters-opus-tier-spacexai-challenges-claude-and-gpt-4-turbo ai|expert en 2026-07-09T17:14:23.900Z Grok 4.5 Launches at $2.49 Per Agentic Task, Undercutting Opus https://aiexpert.news/pt/article/grok-45-enters-opus-tier-spacexai-challenges-claude-and-gpt-4-turbo ai|expert pt-BR 2026-07-09T17:14:23.900Z Grok 4.5 é lançado por R$ 2,49 por tarefa Agentic, cortando a frente do Opus https://aiexpert.news/es/article/grok-45-enters-opus-tier-spacexai-challenges-claude-and-gpt-4-turbo ai|expert es 2026-07-09T17:14:23.900Z Grok 4.5 se lanza a $2.49 por tarea de Agentic, superando a Opus https://aiexpert.news/en/article/cloudflare-urges-immediate-ml-dsa-adoption-amid-quantum-threats ai|expert en 2026-07-09T14:50:24.292Z Cloudflare Pushes ML-DSA Adoption as Quantum Threat Timeline Compresses https://aiexpert.news/pt/article/cloudflare-urges-immediate-ml-dsa-adoption-amid-quantum-threats ai|expert pt-BR 2026-07-09T14:50:24.292Z Cloudflare Impulsiona a Adoção de ML-DSA à medida que Cronograma de Ameaças Quântica se Aproxima https://aiexpert.news/es/article/cloudflare-urges-immediate-ml-dsa-adoption-amid-quantum-threats ai|expert es 2026-07-09T14:50:24.292Z Cloudflare impulsa la adopción de ML-DSA mientras la línea de tiempo de la amenaza cuántica se comprime https://aiexpert.news/en/article/ai-companies-spend-millions-on-elections-to-shape-policy-and-compliance-rules ai|expert en 2026-07-09T11:54:23.549Z AI PACs Spend $200 Million to Rewrite Compliance Rules https://aiexpert.news/pt/article/ai-companies-spend-millions-on-elections-to-shape-policy-and-compliance-rules ai|expert pt-BR 2026-07-09T11:54:23.549Z AI PACs Gastam US$ 200 Milhões para Redigir Regras de Conformidade https://aiexpert.news/es/article/ai-companies-spend-millions-on-elections-to-shape-policy-and-compliance-rules ai|expert es 2026-07-09T11:54:23.549Z IA PACs Gastan $200 Millones para Reescribir Reglas de Cumplimiento https://aiexpert.news/en/article/ai-servers-will-dominate-data-center-power-by-2027-gartner-forecasts-26-growth-t ai|expert en 2026-07-09T11:22:23.402Z AI Servers to Draw 258 Terawatts by 2027, Strangling Grid Capacity https://aiexpert.news/pt/article/ai-servers-will-dominate-data-center-power-by-2027-gartner-forecasts-26-growth-t ai|expert pt-BR 2026-07-09T11:22:23.402Z Servidores de IA vão consumir 258 Terawatts até 2027, sufocando a capacidade da rede elétrica https://aiexpert.news/es/article/ai-servers-will-dominate-data-center-power-by-2027-gartner-forecasts-26-growth-t ai|expert es 2026-07-09T11:22:23.402Z Servidores de IA absorberán 258 Terawatts para 2027, ahogando la capacidad de la red https://aiexpert.news/en/article/openai-applies-epidemiology-to-debug-gnu-libunwind-crashes-at-scale ai|expert en 2026-07-09T10:50:23.449Z OpenAI Found an 18-Year-Old Bug Through Fleet-Scale Data https://aiexpert.news/pt/article/openai-applies-epidemiology-to-debug-gnu-libunwind-crashes-at-scale ai|expert pt-BR 2026-07-09T10:50:23.449Z OpenAI Encontrou um Bug de 18 Anos com Dados em Escala de Frota https://aiexpert.news/es/article/openai-applies-epidemiology-to-debug-gnu-libunwind-crashes-at-scale ai|expert es 2026-07-09T10:50:23.449Z OpenAI Encontró un Error de 18 Años a Través de Datos a Escala de Flota https://aiexpert.news/en/article/alloydb-introduces-local-inferencereplacing-costly-llm-api-calls-with-in-databas ai|expert en 2026-07-09T08:58:23.911Z AlloyDB Trains Local Models Inside PostgreSQL for 23,000× Speed Gains https://aiexpert.news/pt/article/alloydb-introduces-local-inferencereplacing-costly-llm-api-calls-with-in-databas ai|expert pt-BR 2026-07-09T08:58:23.911Z AlloyDB Treina Modelos Locais Dentro do PostgreSQL para Ganhos de Velocidade de 23.000× https://aiexpert.news/es/article/alloydb-introduces-local-inferencereplacing-costly-llm-api-calls-with-in-databas ai|expert es 2026-07-09T08:58:23.911Z AlloyDB Entrena Modelos Locales Dentro de PostgreSQL para Ganancias de Velocidad de 23,000× https://aiexpert.news/en/article/continuous-query-llms-trading-memorization-for-knowledge-base-control ai|expert en 2026-07-09T03:52:23.517Z Cornell's Co-LMLM Matches GPT-4o-Mini by Storing Facts in a Database https://aiexpert.news/pt/article/continuous-query-llms-trading-memorization-for-knowledge-base-control ai|expert pt-BR 2026-07-09T03:52:23.517Z Co-LMLM da Cornell empata GPT-4o-Mini armazenando fatos em um banco de dados https://aiexpert.news/es/article/continuous-query-llms-trading-memorization-for-knowledge-base-control ai|expert es 2026-07-09T03:52:23.517Z Co-LMLM de Cornell empareja a GPT-4o-Mini al almacenar hechos en una base de datos https://aiexpert.news/en/article/selective-timestep-weighting-for-diffusion-rlhf-sample-efficient-alignment-of-im ai|expert en 2026-07-09T03:20:23.561Z Timestep Weighting Cuts Reward-Model Query Costs for Diffusion RLHF https://aiexpert.news/pt/article/selective-timestep-weighting-for-diffusion-rlhf-sample-efficient-alignment-of-im ai|expert pt-BR 2026-07-09T03:20:23.561Z Ponderação de Timestep Reduz Custos de Consulta ao Modelo de Recompensa para RLHF de Difusão https://aiexpert.news/es/article/selective-timestep-weighting-for-diffusion-rlhf-sample-efficient-alignment-of-im ai|expert es 2026-07-09T03:20:23.561Z El Pesaje de Timesteps Reduce el Costo de Consultas al Modelo de Recompensa para RLHF de Difusión https://aiexpert.news/en/article/new-benchmark-tests-multi-agent-ai-deployment-rules ai|expert en 2026-07-09T03:00:23.994Z STRACE Framework Boosts Multi-Agent Verification by 16 Points https://aiexpert.news/pt/article/new-benchmark-tests-multi-agent-ai-deployment-rules ai|expert pt-BR 2026-07-09T03:00:23.994Z STRACE Framework Aumenta Verificação Multi-Agente em 16 Pontos https://aiexpert.news/es/article/new-benchmark-tests-multi-agent-ai-deployment-rules ai|expert es 2026-07-09T03:00:23.994Z El Marco STRACE Aumenta la Verificación Multi-Agente en 16 Puntos https://aiexpert.news/en/article/dynakrag-learnable-multi-hop-rag-control-flow-without-hand-tuned-prompts ai|expert en 2026-07-09T01:30:23.638Z DynaKRAG Boosts Multi-Hop QA Accuracy by Up to 5.78 Points https://aiexpert.news/pt/article/dynakrag-learnable-multi-hop-rag-control-flow-without-hand-tuned-prompts ai|expert pt-BR 2026-07-09T01:30:23.638Z DynaKRAG Aumenta Precisão em QA Multi-Hop até 5,78 Pontos https://aiexpert.news/es/article/dynakrag-learnable-multi-hop-rag-control-flow-without-hand-tuned-prompts ai|expert es 2026-07-09T01:30:23.638Z DynaKRAG Mejora la Precisión de QA Multi-Salto hasta 5.78 Puntos https://aiexpert.news/en/article/modal-cto-on-agent-workload-infrastructure-serverless-meets-stateful-multi-step- ai|expert en 2026-07-09T00:58:23.500Z Agent Workloads Drive Modal's $300 Million Sandbox Revenue https://aiexpert.news/pt/article/modal-cto-on-agent-workload-infrastructure-serverless-meets-stateful-multi-step- ai|expert pt-BR 2026-07-09T00:58:23.500Z Cargas de Agentes Impulsionam Receita de Sandboxes da Modal de US$ 300 Milhões https://aiexpert.news/es/article/modal-cto-on-agent-workload-infrastructure-serverless-meets-stateful-multi-step- ai|expert es 2026-07-09T00:58:23.500Z Cargas de trabajo de agentes impulsan los ingresos de Sandbox de Modal de 300 millones de dólares https://aiexpert.news/en/article/rapidus-sets-2027-deadline-for-new-chip-fab-with-60-customers ai|expert en 2026-07-09T00:26:23.567Z Rapidus Targets 2nm Production by 2027 With No Named Customers https://aiexpert.news/pt/article/rapidus-sets-2027-deadline-for-new-chip-fab-with-60-customers ai|expert pt-BR 2026-07-09T00:26:23.567Z Rapidus Alvo de Produção em 2nm para 2027 Sem Clientes Nomeados https://aiexpert.news/es/article/rapidus-sets-2027-deadline-for-new-chip-fab-with-60-customers ai|expert es 2026-07-09T00:26:23.567Z Rápidus Se Dirige a Producción de 2nm para 2027 Sin Clientes Nombrados