https://aiexpert.news/en/article/google-adds-cycle-level-kernel-profiling-to-xprof ai|expert en 2026-09-23T10:32:38Z Google exposes cycle-level kernel profiling in XProf for Pallas https://aiexpert.news/pt/article/google-adds-cycle-level-kernel-profiling-to-xprof ai|expert pt-BR 2026-09-23T10:32:38Z Google expõe criação de perfil de kernel em nível de ciclo no XProf para Pallas https://aiexpert.news/es/article/google-adds-cycle-level-kernel-profiling-to-xprof ai|expert es 2026-09-23T10:32:38Z Google expone perfilado de kernels a nivel de ciclo en XProf para Pallas https://aiexpert.news/en/article/jev-in-25-lines-of-python ai|expert en 2026-09-23T09:10:48Z Jev choice classification runs in 25 lines of Python with local inference https://aiexpert.news/pt/article/jev-in-25-lines-of-python ai|expert pt-BR 2026-09-23T09:10:48Z Classificação Jev em 25 linhas de Python com inferência local https://aiexpert.news/es/article/jev-in-25-lines-of-python ai|expert es 2026-09-23T09:10:48Z Clasificación de opciones Jev en 25 líneas de Python con inferencia local https://aiexpert.news/en/article/heap-overflow-and-sso-misconfiguration-compromised-openai-internal-repos ai|expert en 2026-09-23T05:52:57Z Heap overflow in libheif let researchers breach OpenAI in 72 hours https://aiexpert.news/pt/article/heap-overflow-and-sso-misconfiguration-compromised-openai-internal-repos ai|expert pt-BR 2026-09-23T05:52:57Z Heap overflow em libheif permitiu pesquisadores brechar OpenAI em 72 horas https://aiexpert.news/es/article/heap-overflow-and-sso-misconfiguration-compromised-openai-internal-repos ai|expert es 2026-09-23T05:52:57Z Desbordamiento de heap en libheif permitió a investigadores comprometer OpenAI en 72 hours https://aiexpert.news/en/article/the-agent-harness-control-planes-invariants-and-approval-boundaries-for-producti ai|expert en 2026-09-23T05:52:31Z OpenAI's Govindarajan: agent reliability lives in the harness, not the model https://aiexpert.news/pt/article/the-agent-harness-control-planes-invariants-and-approval-boundaries-for-producti ai|expert pt-BR 2026-09-23T05:52:31Z OpenAI's Govindarajan: agent reliability lives in the harness, not the model https://aiexpert.news/es/article/the-agent-harness-control-planes-invariants-and-approval-boundaries-for-producti ai|expert es 2026-09-23T05:52:31Z OpenAI's Govindarajan: agent reliability lives in the harness, not the model https://aiexpert.news/en/article/netflix-maps-multimodal-asset-personalization-at-scale ai|expert en 2026-09-23T05:52:19Z Netflix unified artwork ranking across five canvases with CLIP embeddings https://aiexpert.news/pt/article/netflix-maps-multimodal-asset-personalization-at-scale ai|expert pt-BR 2026-09-23T05:52:19Z Netflix unifica ranking de capas em cinco canvases com embeddings CLIP https://aiexpert.news/es/article/netflix-maps-multimodal-asset-personalization-at-scale ai|expert es 2026-09-23T05:52:19Z Netflix unifica ranking de artwork en cinco canvas con embeddings CLIP https://aiexpert.news/en/article/nvidia-announces-native-gpu-programming-in-rust-with-two-kernel-writing-tracks ai|expert en 2026-09-23T05:51:06Z Nvidia brings native GPU kernel programming to Rust with cuda-oxide and cutile-rs https://aiexpert.news/pt/article/nvidia-announces-native-gpu-programming-in-rust-with-two-kernel-writing-tracks ai|expert pt-BR 2026-09-23T05:51:06Z Nvidia traz programação nativa de kernels GPU para Rust com cuda-oxide e cutile-rs https://aiexpert.news/es/article/nvidia-announces-native-gpu-programming-in-rust-with-two-kernel-writing-tracks ai|expert es 2026-09-23T05:51:06Z NVIDIA trae programación nativa de kernels GPU a Rust con cuda-oxide y cutile-rs https://aiexpert.news/en/article/cloudflare-saved-100tb-of-ram-through-algorithmic-optimization-in-rust ai|expert en 2026-09-23T05:42:10Z Cloudflare reclaimed 100TB of RAM by cutting hash points 90% https://aiexpert.news/pt/article/cloudflare-saved-100tb-of-ram-through-algorithmic-optimization-in-rust ai|expert pt-BR 2026-09-23T05:42:10Z Cloudflare recuperou 100TB de RAM cortando pontos de hash em 90% https://aiexpert.news/es/article/cloudflare-saved-100tb-of-ram-through-algorithmic-optimization-in-rust ai|expert es 2026-09-23T05:42:10Z Cloudflare recuperó 100TB de RAM al reducir puntos hash 90% https://aiexpert.news/en/article/can-ai-design-circuit-boards-yet ai|expert en 2026-09-23T05:42:09Z Claude Opus 5 leads circuit design benchmark at 61.6%, still fails 4 in 10 https://aiexpert.news/pt/article/can-ai-design-circuit-boards-yet ai|expert pt-BR 2026-09-23T05:42:09Z Claude Opus 5 lidera benchmark de design de circuitos com 61.6%, mas falha em 4 de 10 https://aiexpert.news/es/article/can-ai-design-circuit-boards-yet ai|expert es 2026-09-23T05:42:09Z Claude Opus 5 lidera benchmark de diseño de circuitos con 61.6%, pero falla en 4 de 10 https://aiexpert.news/en/article/cognition-launches-swe-2-model-for-code-generation ai|expert en 2026-09-23T05:32:47Z Cognition's SWE-2 matches frontier code performance at 64% lower cost https://aiexpert.news/pt/article/cognition-launches-swe-2-model-for-code-generation ai|expert pt-BR 2026-09-23T05:32:47Z SWE-2 da Cognition iguala desempenho de código frontier com 64% menos custo https://aiexpert.news/es/article/cognition-launches-swe-2-model-for-code-generation ai|expert es 2026-09-23T05:32:47Z SWE-2 de Cognition iguala rendimiento de código frontier a 64% menor costo https://aiexpert.news/en/article/openai-agents-api ai|expert en 2026-09-23T05:32:09Z OpenAI Agents API limits data residency to United States https://aiexpert.news/pt/article/openai-agents-api ai|expert pt-BR 2026-09-23T05:32:09Z API de Agentes da OpenAI limita residência de dados aos Estados Unidos https://aiexpert.news/es/article/openai-agents-api ai|expert es 2026-09-23T05:32:09Z API de Agentes de OpenAI limita data residency a Estados Unidos https://aiexpert.news/en/article/knowledge-pull-requests-integrate-new-knowledge-into-documents-while-preserving- ai|expert en 2026-09-23T05:32:04Z Knowledge Pull Requests ground QA 69% on revised articles versus 48% for raw rewrites https://aiexpert.news/pt/article/knowledge-pull-requests-integrate-new-knowledge-into-documents-while-preserving- ai|expert pt-BR 2026-09-23T05:32:04Z Knowledge Pull Requests alcançam 69% de acurácia em QA versus 48% para reescritas brutas https://aiexpert.news/es/article/knowledge-pull-requests-integrate-new-knowledge-into-documents-while-preserving- ai|expert es 2026-09-23T05:32:04Z Knowledge Pull Requests logran 69% en QA sobre artículos revisados versus 48% en reescrituras brutas https://aiexpert.news/en/article/gravityocr-394-decode-speedup-on-document-ocr-via-ar-diffusion-drafting ai|expert en 2026-09-23T05:22:45Z GravityOCR achieves 3.94× document OCR speedup via speculative decoding https://aiexpert.news/pt/article/gravityocr-394-decode-speedup-on-document-ocr-via-ar-diffusion-drafting ai|expert pt-BR 2026-09-23T05:22:45Z GravityOCR alcança 3.94× de aceleração em OCR de documentos via decodificação especulativa https://aiexpert.news/es/article/gravityocr-394-decode-speedup-on-document-ocr-via-ar-diffusion-drafting ai|expert es 2026-09-23T05:22:45Z GravityOCR logra aceleración de 3.94× en OCR de documentos mediante decodificación especulativa https://aiexpert.news/en/article/frontier-models-externalize-token-efficient-reasoning-gpt-6-astra-selects-correc ai|expert en 2026-09-23T05:22:00Z GPT-6 Astra's compressed reasoning traces don't transfer to weaker models https://aiexpert.news/pt/article/frontier-models-externalize-token-efficient-reasoning-gpt-6-astra-selects-correc ai|expert pt-BR 2026-09-23T05:22:00Z Os traces de raciocínio comprimido do GPT-6 Astra não se transferem para modelos mais fracos https://aiexpert.news/es/article/frontier-models-externalize-token-efficient-reasoning-gpt-6-astra-selects-correc ai|expert es 2026-09-23T05:22:00Z Los rastros de razonamiento comprimido de GPT-6 Astra no se transfieren a modelos más débiles https://aiexpert.news/en/article/ai-power-demands-push-gan-into-data-center-design ai|expert en 2026-09-23T05:21:27Z Silicon power conversion hits ceiling as AI GPUs demand 5 kW within two years https://aiexpert.news/pt/article/ai-power-demands-push-gan-into-data-center-design ai|expert pt-BR 2026-09-23T05:21:27Z Conversão de potência em silício atinge limite com GPUs de IA demandando 5 kW em dois anos https://aiexpert.news/es/article/ai-power-demands-push-gan-into-data-center-design ai|expert es 2026-09-23T05:21:27Z Silicon power conversion hits ceiling as AI GPUs demand 5 kW within two years https://aiexpert.news/en/article/greedy-decoding-diverges-49100-across-bf16-and-fp16-selective-fp32-lm-head-recom ai|expert en 2026-09-23T05:21:10Z Greedy decoding diverges 49–100% across BF16 and FP16; selective FP32 head recomputation recovers +22–36 pp https://aiexpert.news/pt/article/greedy-decoding-diverges-49100-across-bf16-and-fp16-selective-fp32-lm-head-recom ai|expert pt-BR 2026-09-23T05:21:10Z Decodificação gulosa diverge 49–100% entre BF16 e FP16; recomputação seletiva FP32 da cabeça recupera +22–36 pp https://aiexpert.news/es/article/greedy-decoding-diverges-49100-across-bf16-and-fp16-selective-fp32-lm-head-recom ai|expert es 2026-09-23T05:21:10Z Decodificación greedy diverge 49–100% entre BF16 y FP16; recomputación selectiva FP32 recupera +22–36 pp https://aiexpert.news/en/article/agensh-scaling-organizational-intelligence-to-1024-agents ai|expert en 2026-09-23T05:02:22Z Microsoft's Agensh reaches 55% test-pass rate with 1,024 agents https://aiexpert.news/pt/article/agensh-scaling-organizational-intelligence-to-1024-agents ai|expert pt-BR 2026-09-23T05:02:22Z Agensh da Microsoft atinge 55% de taxa de aprovação com 1.024 agentes https://aiexpert.news/es/article/agensh-scaling-organizational-intelligence-to-1024-agents ai|expert es 2026-09-23T05:02:22Z Agensh de Microsoft alcanza 55% de tasa de aprobación con 1,024 agentes https://aiexpert.news/en/article/swe-serve-benchmarking-agentic-engineering-for-production-inference-serving ai|expert en 2026-09-23T05:02:09Z SWE-Serve benchmark reveals 23-point gap between local tests and production serving https://aiexpert.news/pt/article/swe-serve-benchmarking-agentic-engineering-for-production-inference-serving ai|expert pt-BR 2026-09-23T05:02:09Z SWE-Serve revela 23-point gap entre testes locais e serving em produção https://aiexpert.news/es/article/swe-serve-benchmarking-agentic-engineering-for-production-inference-serving ai|expert es 2026-09-23T05:02:09Z SWE-Serve benchmark expone 23-point gap entre pruebas locales y serving en producción https://aiexpert.news/en/article/a2m-trace-optimized-agent-hijacking-in-the-mcp-ecosystem ai|expert en 2026-09-23T04:50:55Z A2M attack hijacks MCP agents with poisoned tool metadata at 93.6% success https://aiexpert.news/pt/article/a2m-trace-optimized-agent-hijacking-in-the-mcp-ecosystem ai|expert pt-BR 2026-09-23T04:50:55Z Ataque A2M sequestra agentes MCP com metadados de ferramentas envenenadas a 93.6% https://aiexpert.news/es/article/a2m-trace-optimized-agent-hijacking-in-the-mcp-ecosystem ai|expert es 2026-09-23T04:50:55Z Ataque A2M secuestra agentes MCP con metadatos de herramientas envenenadas al 93.6% https://aiexpert.news/en/article/harnesstax-how-much-does-the-harness-matter-for-coding-agents ai|expert en 2026-09-23T04:21:50Z Harness choice drives 5x cost gap while coding agent success rates stay flat https://aiexpert.news/pt/article/harnesstax-how-much-does-the-harness-matter-for-coding-agents ai|expert pt-BR 2026-09-23T04:21:50Z Escolha de harness gera diferença de custo de 5x enquanto taxas de sucesso de agentes de código permanecem planas https://aiexpert.news/es/article/harnesstax-how-much-does-the-harness-matter-for-coding-agents ai|expert es 2026-09-23T04:21:50Z La elección de harness genera una brecha de costo de 5x mientras las tasas de éxito de agentes de código se mantienen planas https://aiexpert.news/en/article/gitlab-duo-expands-self-hosted-ai-options-through-microsoft-foundry ai|expert en 2026-09-23T04:20:38Z GitLab Duo adds self-hosted models through Microsoft Foundry https://aiexpert.news/pt/article/gitlab-duo-expands-self-hosted-ai-options-through-microsoft-foundry ai|expert pt-BR 2026-09-23T04:20:38Z GitLab Duo adiciona modelos auto-hospedados via Microsoft Foundry https://aiexpert.news/es/article/gitlab-duo-expands-self-hosted-ai-options-through-microsoft-foundry ai|expert es 2026-09-23T04:20:38Z GitLab Duo añade modelos auto-hospedados a través de Microsoft Foundry https://aiexpert.news/en/article/flash-dllm-io-aware-kv-caching-and-parallel-decoding-for-diffusion-llms ai|expert en 2026-09-23T04:10:46Z Flash-dLLM cuts diffusion LLM inference latency 11× with fused kernels https://aiexpert.news/pt/article/flash-dllm-io-aware-kv-caching-and-parallel-decoding-for-diffusion-llms ai|expert pt-BR 2026-09-23T04:10:46Z Flash-dLLM cuts diffusion LLM inference latency 11× with fused kernels https://aiexpert.news/es/article/flash-dllm-io-aware-kv-caching-and-parallel-decoding-for-diffusion-llms ai|expert es 2026-09-23T04:10:46Z Flash-dLLM cuts diffusion LLM inference latency 11× with fused kernels https://aiexpert.news/en/article/ai-coding-made-ci-a-bottleneck-linear-reworked-theirs-to-keep-up ai|expert en 2026-09-23T03:41:08Z Linear halved CI runner time despite test suite quadrupling https://aiexpert.news/pt/article/ai-coding-made-ci-a-bottleneck-linear-reworked-theirs-to-keep-up ai|expert pt-BR 2026-09-23T03:41:08Z Linear reduziu tempo de CI runner pela metade apesar de suite de testes quadruplicar https://aiexpert.news/es/article/ai-coding-made-ci-a-bottleneck-linear-reworked-theirs-to-keep-up ai|expert es 2026-09-23T03:41:08Z Linear redujo a la mitad el tiempo del runner de CI a pesar de que la suite de pruebas se cuadruplicó https://aiexpert.news/en/article/cloudflare-worker-previews-isolated-environments-for-agent-changes ai|expert en 2026-09-23T03:40:49Z Cloudflare Worker Previews isolate agent changes in production-like branches https://aiexpert.news/pt/article/cloudflare-worker-previews-isolated-environments-for-agent-changes ai|expert pt-BR 2026-09-23T03:40:49Z Cloudflare Worker Previews isolam mudanças de agentes em branches similares à produção https://aiexpert.news/es/article/cloudflare-worker-previews-isolated-environments-for-agent-changes ai|expert es 2026-09-23T03:40:49Z Cloudflare Worker Previews aíslan cambios de agentes en ramas similares a producción https://aiexpert.news/en/article/databricks-genie-one-mcp-reaches-general-availability-for-agent-business-context ai|expert en 2026-09-23T03:25:31Z Databricks Genie One MCP reaches general availability https://aiexpert.news/pt/article/databricks-genie-one-mcp-reaches-general-availability-for-agent-business-context ai|expert pt-BR 2026-09-23T03:25:31Z Databricks Genie One MCP chega à disponibilidade geral https://aiexpert.news/es/article/databricks-genie-one-mcp-reaches-general-availability-for-agent-business-context ai|expert es 2026-09-23T03:25:31Z Databricks Genie One MCP llega a disponibilidad general