https://aiexpert.news/en/article/equilibrium-reasoners-test-time-compute-scaling-via-attractor-learning ai|expert en 2026-05-21T11:48:15.401Z Equilibrium Reasoners lift Sudoku accuracy from 2.6% to 99% via test-time scaling https://aiexpert.news/pt/article/equilibrium-reasoners-test-time-compute-scaling-via-attractor-learning ai|expert pt-BR 2026-05-21T11:48:15.401Z Equilibrium Reasoners elevam precisão do Sudoku de 2,6% para 99% via test-time scaling https://aiexpert.news/es/article/equilibrium-reasoners-test-time-compute-scaling-via-attractor-learning ai|expert es 2026-05-21T11:48:15.401Z Equilibrium Reasoners elevan la precisión del Sudoku de 2,6% a 99% mediante test-time scaling https://aiexpert.news/en/article/carv-reduces-diffusion-teacher-variance-for-cheaper-3d-distillation ai|expert en 2026-05-21T08:54:15.406Z NVIDIA's CARV cuts 3D distillation compute by 2–3× https://aiexpert.news/pt/article/carv-reduces-diffusion-teacher-variance-for-cheaper-3d-distillation ai|expert pt-BR 2026-05-21T08:54:15.406Z CARV da NVIDIA reduz computação de destilação 3D em 2–3× https://aiexpert.news/es/article/carv-reduces-diffusion-teacher-variance-for-cheaper-3d-distillation ai|expert es 2026-05-21T08:54:15.406Z CARV de NVIDIA reduce el cómputo de destilación 3D en 2–3× https://aiexpert.news/en/article/hyperparameter-transfer-at-scale-embedding-layer-learning-rate-matters ai|expert en 2026-05-21T05:08:15.269Z One hyperparameter rule captures most of µP's gains https://aiexpert.news/pt/article/hyperparameter-transfer-at-scale-embedding-layer-learning-rate-matters ai|expert pt-BR 2026-05-21T05:08:15.269Z Uma regra de hiperparâmetro captura a maioria dos ganhos de μP https://aiexpert.news/es/article/hyperparameter-transfer-at-scale-embedding-layer-learning-rate-matters ai|expert es 2026-05-21T05:08:15.269Z Una regla de hiperparámetro captura la mayoría de las ganancias de μP https://aiexpert.news/en/article/rlvr-weight-updates-follow-rank-1-patterns-fewer-tokens-matter ai|expert en 2026-05-21T04:36:15.362Z RELEX reconstructs RLVR checkpoints from 15% training data https://aiexpert.news/pt/article/rlvr-weight-updates-follow-rank-1-patterns-fewer-tokens-matter ai|expert pt-BR 2026-05-21T04:36:15.362Z RELEX reconstrói checkpoints RLVR a partir de 15% dos dados de treinamento https://aiexpert.news/es/article/rlvr-weight-updates-follow-rank-1-patterns-fewer-tokens-matter ai|expert es 2026-05-21T04:36:15.362Z RELEX reconstruye checkpoints RLVR a partir del 15% de los datos de entrenamiento https://aiexpert.news/en/article/deepweb-bench-a-harder-eval-for-frontier-agent-research ai|expert en 2026-05-21T04:04:15.366Z Peking researchers release DeepWeb-Bench, exposing derivation failures in frontier AI https://aiexpert.news/pt/article/deepweb-bench-a-harder-eval-for-frontier-agent-research ai|expert pt-BR 2026-05-21T04:04:15.366Z Pesquisadores de Pequim divulgam DeepWeb-Bench, expondo falhas de derivação em IA de fronteira https://aiexpert.news/es/article/deepweb-bench-a-harder-eval-for-frontier-agent-research ai|expert es 2026-05-21T04:04:15.366Z Investigadores de Peking divulgan DeepWeb-Bench, exponiendo fallos de derivación en IA de frontera https://aiexpert.news/en/article/agent-jit-compilation-cuts-latency-in-tool-use-workflows ai|expert en 2026-05-21T03:32:15.261Z Agent JIT compilation cuts latency 10.4× over Browser-Use https://aiexpert.news/pt/article/agent-jit-compilation-cuts-latency-in-tool-use-workflows ai|expert pt-BR 2026-05-21T03:32:15.261Z Compilação JIT de agentes reduz latência 10.4× em relação a Browser-Use https://aiexpert.news/es/article/agent-jit-compilation-cuts-latency-in-tool-use-workflows ai|expert es 2026-05-21T03:32:15.261Z Compilación JIT de agentes reduce latencia 10.4× versus Browser-Use https://aiexpert.news/en/article/reasoning-trace-collapse-fine-tuning-wipes-out-explicit-reasoning-chains ai|expert en 2026-05-21T03:00:55.248Z Fine-tuning erases reasoning chains while accuracy stays high https://aiexpert.news/pt/article/reasoning-trace-collapse-fine-tuning-wipes-out-explicit-reasoning-chains ai|expert pt-BR 2026-05-21T03:00:55.248Z Fine-tuning apaga cadeias de raciocínio enquanto precisão se mantém alta https://aiexpert.news/es/article/reasoning-trace-collapse-fine-tuning-wipes-out-explicit-reasoning-chains ai|expert es 2026-05-21T03:00:55.248Z Fine-tuning borra cadenas de razonamiento mientras la precisión se mantiene alta https://aiexpert.news/en/article/nvidia-concedes-china-ai-chip-market-to-huawei ai|expert en 2026-05-21T02:30:15.301Z Nvidia cedes China AI chip market to Huawei https://aiexpert.news/pt/article/nvidia-concedes-china-ai-chip-market-to-huawei ai|expert pt-BR 2026-05-21T02:30:15.301Z Nvidia cede mercado de chips de IA na China para Huawei https://aiexpert.news/es/article/nvidia-concedes-china-ai-chip-market-to-huawei ai|expert es 2026-05-21T02:30:15.301Z Nvidia cede el mercado de chips de IA en China a Huawei https://aiexpert.news/en/article/opencomputer-verifiable-software-worlds-for-agent-evaluation ai|expert en 2026-05-20T02:15:56.642Z OpenComputer Replaces LLM Judges With Verifiable Desktop Tasks https://aiexpert.news/pt/article/opencomputer-verifiable-software-worlds-for-agent-evaluation ai|expert pt-BR 2026-05-20T02:15:56.642Z OpenComputer Substitui Juízes de LLM por Tarefas de Desktop Verificáveis https://aiexpert.news/es/article/opencomputer-verifiable-software-worlds-for-agent-evaluation ai|expert es 2026-05-20T02:15:56.642Z OpenComputer Reemplaza Jueces de LLM por Tareas de Desktop Verificables https://aiexpert.news/en/article/gemini-35-flash-launches-at-3-the-price-of-its-predecessor-and-google-is-deployi ai|expert en 2026-05-19T23:18:15.619Z Gemini 3.5 Flash Launches Globally at 3× Cost https://aiexpert.news/pt/article/gemini-35-flash-launches-at-3-the-price-of-its-predecessor-and-google-is-deployi ai|expert pt-BR 2026-05-19T23:18:15.619Z Gemini 3.5 Flash é Lançado Globalmente a 3× de Custo https://aiexpert.news/es/article/gemini-35-flash-launches-at-3-the-price-of-its-predecessor-and-google-is-deployi ai|expert es 2026-05-19T23:18:15.619Z Gemini 3.5 Flash se Lanza Globalmente a 3× de Costo https://aiexpert.news/en/article/allen-ais-olmoearth-v11-cuts-satellite-imagery-inference-costs-by-3x-with-smarte ai|expert en 2026-05-19T20:50:15.661Z Allen AI's OlmoEarth v1.1 cuts satellite inference compute 3x https://aiexpert.news/pt/article/allen-ais-olmoearth-v11-cuts-satellite-imagery-inference-costs-by-3x-with-smarte ai|expert pt-BR 2026-05-19T20:50:15.661Z OlmoEarth v1.1 da Allen AI reduz computação em inferência de satélite em 3x https://aiexpert.news/es/article/allen-ais-olmoearth-v11-cuts-satellite-imagery-inference-costs-by-3x-with-smarte ai|expert es 2026-05-19T20:50:15.661Z OlmoEarth v1.1 de Allen AI reduce computación en inferencia satelital en 3x https://aiexpert.news/en/article/anthropic-mcp-tunnels-let-agents-access-internal-systems-securely ai|expert en 2026-05-19T19:50:16.214Z Anthropic Ships Self-Hosted Sandboxes and MCP Tunnels https://aiexpert.news/pt/article/anthropic-mcp-tunnels-let-agents-access-internal-systems-securely ai|expert pt-BR 2026-05-19T19:50:16.214Z Anthropic Lança Sandboxes Auto-Hospedados e Túneis MCP https://aiexpert.news/es/article/anthropic-mcp-tunnels-let-agents-access-internal-systems-securely ai|expert es 2026-05-19T19:50:16.214Z Anthropic Lanza Sandboxes Auto-Hospedados y Túneles MCP https://aiexpert.news/en/article/chinese-fab-leaders-push-domestic-chipmaking-tools-onto-production-lines ai|expert en 2026-05-19T17:50:16.597Z Chinese Fabs Move Domestic Tools to Production Lines https://aiexpert.news/pt/article/chinese-fab-leaders-push-domestic-chipmaking-tools-onto-production-lines ai|expert pt-BR 2026-05-19T17:50:16.597Z Fábricas Chinesas Transferem Ferramentas Domésticas para Linhas de Produção https://aiexpert.news/es/article/chinese-fab-leaders-push-domestic-chipmaking-tools-onto-production-lines ai|expert es 2026-05-19T17:50:16.597Z Fábricas Chinas Trasladan Herramientas Domésticas a Líneas de Producción