https://aiexpert.news/en/article/psiquantums-8x8-photonic-switch-cuts-optical-loss-for-ai-gpu-networks ai|expert en 2026-10-30T15:15:10Z PsiQuantum's Photonic Switch Loses Just 1.2 dB, Below Nvidia's 3 dB Cap https://aiexpert.news/pt/article/psiquantums-8x8-photonic-switch-cuts-optical-loss-for-ai-gpu-networks ai|expert pt-BR 2026-10-30T15:15:10Z Switch fotônico da PsiQuantum perde apenas 1.2 dB, abaixo do limite de 3 dB da Nvidia https://aiexpert.news/es/article/psiquantums-8x8-photonic-switch-cuts-optical-loss-for-ai-gpu-networks ai|expert es 2026-10-30T15:15:10Z El conmutador fotónico de PsiQuantum pierde solo 1.2 dB, por debajo del límite de 3 dB de Nvidia https://aiexpert.news/en/article/openai-details-how-it-detects-misalignment-in-frontier-model-behavior ai|expert en 2026-10-30T15:13:58Z OpenAI details a four-layer system for catching misaligned frontier agents https://aiexpert.news/pt/article/openai-details-how-it-detects-misalignment-in-frontier-model-behavior ai|expert pt-BR 2026-10-30T15:13:58Z OpenAI detalha sistema de quatro camadas para detectar agentes de fronteira desalinhados https://aiexpert.news/es/article/openai-details-how-it-detects-misalignment-in-frontier-model-behavior ai|expert es 2026-10-30T15:13:58Z OpenAI detalla un sistema de cuatro capas para detectar agentes de frontera desalineados https://aiexpert.news/en/article/paperpilot-ships-a-zero-cloud-rust-pdf-engine-with-wasm-and-mcp-interfaces ai|expert en 2026-10-30T15:13:18Z PaperPilot ships a zero-cloud, Rust PDF engine with WASM and MCP interfaces https://aiexpert.news/pt/article/paperpilot-ships-a-zero-cloud-rust-pdf-engine-with-wasm-and-mcp-interfaces ai|expert pt-BR 2026-10-30T15:13:18Z PaperPilot lança motor PDF em Rust zero-cloud com interfaces WASM e MCP https://aiexpert.news/es/article/paperpilot-ships-a-zero-cloud-rust-pdf-engine-with-wasm-and-mcp-interfaces ai|expert es 2026-10-30T15:13:18Z PaperPilot lanza un motor PDF en Rust zero-cloud con interfaces WASM y MCP https://aiexpert.news/en/article/doordash-cuts-grocery-reorder-latency-from-16-seconds-to-2-seconds-with-ai-agent ai|expert en 2026-10-30T15:12:57Z DoorDash cuts grocery reorder latency from 16 seconds to 2 seconds with AI agents https://aiexpert.news/pt/article/doordash-cuts-grocery-reorder-latency-from-16-seconds-to-2-seconds-with-ai-agent ai|expert pt-BR 2026-10-30T15:12:57Z DoorDash reduz latência de recompra de mercado de 16 segundos para 2 segundos com agentes de IA https://aiexpert.news/es/article/doordash-cuts-grocery-reorder-latency-from-16-seconds-to-2-seconds-with-ai-agent ai|expert es 2026-10-30T15:12:57Z DoorDash reduce la latencia de reórdenes de supermercado de 16 segundos a 2 segundos con agentes de IA https://aiexpert.news/en/article/cloudflare-discloses-cross-request-kv-cache-exposure-in-glm-53-on-workers-ai ai|expert en 2026-10-30T15:11:58Z Cloudflare says GLM-5.3 bug on Workers AI leaked data in 155 responses https://aiexpert.news/pt/article/cloudflare-discloses-cross-request-kv-cache-exposure-in-glm-53-on-workers-ai ai|expert pt-BR 2026-10-30T15:11:58Z Cloudflare diz que bug do GLM-5.3 no Workers AI vazou dados em 155 respostas https://aiexpert.news/es/article/cloudflare-discloses-cross-request-kv-cache-exposure-in-glm-53-on-workers-ai ai|expert es 2026-10-30T15:11:58Z Cloudflare dice que un bug de GLM-5.3 en Workers AI filtró datos en 155 respuestas https://aiexpert.news/en/article/netflix-reframes-gpu-reliability-as-a-distributed-systems-engineering-problem ai|expert en 2026-10-30T15:11:42Z Netflix lifts GPU job success to 99.5% with distributed-systems fixes https://aiexpert.news/pt/article/netflix-reframes-gpu-reliability-as-a-distributed-systems-engineering-problem ai|expert pt-BR 2026-10-30T15:11:42Z Netflix eleva taxa de sucesso de jobs em GPU a 99.5% com correções de sistemas distribuídos https://aiexpert.news/es/article/netflix-reframes-gpu-reliability-as-a-distributed-systems-engineering-problem ai|expert es 2026-10-30T15:11:42Z Netflix eleva el éxito de trabajos GPU al 99.5% con correcciones de sistemas distribuidos https://aiexpert.news/en/article/a-single-1b-decoder-replaces-gate-plus-generator-pipelines-for-proactive-assista ai|expert en 2026-10-30T15:10:24Z A single 1B decoder replaces gate-plus-generator pipelines for proactive assistant hints https://aiexpert.news/pt/article/a-single-1b-decoder-replaces-gate-plus-generator-pipelines-for-proactive-assista ai|expert pt-BR 2026-10-30T15:10:24Z Um único decoder de 1B substitui pipelines gate-plus-generator em dicas proativas de assistente https://aiexpert.news/es/article/a-single-1b-decoder-replaces-gate-plus-generator-pipelines-for-proactive-assista ai|expert es 2026-10-30T15:10:24Z Un solo decoder de 1B reemplaza pipelines de gate-y-generador para sugerencias proactivas https://aiexpert.news/en/article/stanford-professors-say-anthropic-trained-claude-on-their-pirated-books ai|expert en 2026-10-30T15:09:52Z Stanford professors say Anthropic trained Claude on their pirated books https://aiexpert.news/pt/article/stanford-professors-say-anthropic-trained-claude-on-their-pirated-books ai|expert pt-BR 2026-10-30T15:09:52Z Professores de Stanford dizem que Anthropic treinou Claude com seus livros piratas https://aiexpert.news/es/article/stanford-professors-say-anthropic-trained-claude-on-their-pirated-books ai|expert es 2026-10-30T15:09:52Z Profesores de Stanford dicen que Anthropic entrenó a Claude con sus libros pirateados https://aiexpert.news/en/article/claude-code-sends-each-background-agents-full-transcript-back-to-the-model ai|expert en 2026-10-30T15:07:13Z Claude Code resends each background agent's transcript every 30 seconds https://aiexpert.news/pt/article/claude-code-sends-each-background-agents-full-transcript-back-to-the-model ai|expert pt-BR 2026-10-30T15:07:13Z Claude Code reenvia a transcrição de cada agente em segundo plano a cada 30 segundos https://aiexpert.news/es/article/claude-code-sends-each-background-agents-full-transcript-back-to-the-model ai|expert es 2026-10-30T15:07:13Z Claude Code reenvía la transcripción de cada agente en segundo plano cada 30 segundos https://aiexpert.news/en/article/time-based-calibration-exposes-which-diffusion-model-images-were-memorized-code- ai|expert en 2026-10-30T15:07:05Z TSC doubles diffusion-model memorization detection rate to 62.3% TPR https://aiexpert.news/pt/article/time-based-calibration-exposes-which-diffusion-model-images-were-memorized-code- ai|expert pt-BR 2026-10-30T15:07:05Z TSC dobra taxa de detecção de memorização em modelos de difusão para 62.3% TPR https://aiexpert.news/es/article/time-based-calibration-exposes-which-diffusion-model-images-were-memorized-code- ai|expert es 2026-10-30T15:07:05Z TSC duplica la tasa de detección de memorización en modelos de difusión al 62.3% TPR https://aiexpert.news/en/article/textseal2-keeps-llm-watermark-signal-intact-under-speculative-decoding-no-added- ai|expert en 2026-10-30T15:06:48Z TextSeal2 keeps LLM watermark signal intact under speculative decoding, no added latency https://aiexpert.news/pt/article/textseal2-keeps-llm-watermark-signal-intact-under-speculative-decoding-no-added- ai|expert pt-BR 2026-10-30T15:06:48Z TextSeal2 mantém sinal de watermark de LLM intacto sob decodificação especulativa, sem latência adicional https://aiexpert.news/es/article/textseal2-keeps-llm-watermark-signal-intact-under-speculative-decoding-no-added- ai|expert es 2026-10-30T15:06:48Z TextSeal2 mantiene intacta la señal de marca de agua en LLM bajo speculative decoding, sin latencia extra https://aiexpert.news/en/article/openai-and-anthropic-slashed-llm-api-prices-by-up-to-80-within-months ai|expert en 2026-10-30T15:05:26Z Cache-read pricing, not list prices, drove the 70-75% LLM cost cuts https://aiexpert.news/pt/article/openai-and-anthropic-slashed-llm-api-prices-by-up-to-80-within-months ai|expert pt-BR 2026-10-30T15:05:26Z Preço de cache-read, não o preço de tabela, gerou os cortes de 70-75% no custo de LLMs https://aiexpert.news/es/article/openai-and-anthropic-slashed-llm-api-prices-by-up-to-80-within-months ai|expert es 2026-10-30T15:05:26Z El precio del cache-read, no las tarifas de lista, impulsó los recortes del 70-75% en costos de LLM https://aiexpert.news/en/article/shallowdefend-trains-llms-to-catch-multi-turn-jailbreaks-at-the-first-reply-turn ai|expert en 2026-10-30T15:05:06Z ShallowDefend lifts multi-turn jailbreak safety to 90.4% without a judge model https://aiexpert.news/pt/article/shallowdefend-trains-llms-to-catch-multi-turn-jailbreaks-at-the-first-reply-turn ai|expert pt-BR 2026-10-30T15:05:06Z ShallowDefend eleva a segurança contra jailbreak multi-turn a 90.4% sem modelo juiz https://aiexpert.news/es/article/shallowdefend-trains-llms-to-catch-multi-turn-jailbreaks-at-the-first-reply-turn ai|expert es 2026-10-30T15:05:06Z ShallowDefend eleva la seguridad ante jailbreaks multi-turno a 90.4% sin modelo juez https://aiexpert.news/en/article/trojanized-claude-code-extension-distributed-through-the-cursor-marketplace ai|expert en 2026-10-30T15:04:56Z Backdoored Claude Code extension hit 5,000 Cursor developer systems https://aiexpert.news/pt/article/trojanized-claude-code-extension-distributed-through-the-cursor-marketplace ai|expert pt-BR 2026-10-30T15:04:56Z Extensão comprometida do Claude Code atingiu 5,000 sistemas de devs no Cursor https://aiexpert.news/es/article/trojanized-claude-code-extension-distributed-through-the-cursor-marketplace ai|expert es 2026-10-30T15:04:56Z Extensión Claude Code con backdoor afectó a 5,000 sistemas de desarrolladores en Cursor https://aiexpert.news/en/article/cerebras-details-how-it-deploys-trillion-parameter-frontier-models-in-production ai|expert en 2026-10-30T15:04:38Z Cerebras splits 3-trillion-parameter Kimi K3 across 59 wafers https://aiexpert.news/pt/article/cerebras-details-how-it-deploys-trillion-parameter-frontier-models-in-production ai|expert pt-BR 2026-10-30T15:04:38Z Cerebras divide Kimi K3 de 3 trilhões de parâmetros em 59 wafers https://aiexpert.news/es/article/cerebras-details-how-it-deploys-trillion-parameter-frontier-models-in-production ai|expert es 2026-10-30T15:04:38Z Cerebras divide a Kimi K3, de 3 billones de parámetros, en 59 wafers https://aiexpert.news/en/article/backdoors-found-in-official-claude-code-extensions-researcher-reports ai|expert en 2026-10-30T15:04:36Z Anthropic's Claude Code patch fails to remove active SSH backdoor https://aiexpert.news/pt/article/backdoors-found-in-official-claude-code-extensions-researcher-reports ai|expert pt-BR 2026-10-30T15:04:36Z Patch do Claude Code, da Anthropic, não remove backdoor SSH ativo https://aiexpert.news/es/article/backdoors-found-in-official-claude-code-extensions-researcher-reports ai|expert es 2026-10-30T15:04:36Z El parche de Claude Code de Anthropic no elimina la puerta trasera SSH activa https://aiexpert.news/en/article/mistral-releases-open-weights-for-mistral-large-4 ai|expert en 2026-10-30T15:04:14Z Mistral Large 4 matches GLM-5.3 on coding while using 47% fewer tokens https://aiexpert.news/pt/article/mistral-releases-open-weights-for-mistral-large-4 ai|expert pt-BR 2026-10-30T15:04:14Z Mistral Large 4 empata com GLM-5.3 em coding com 47% menos tokens https://aiexpert.news/es/article/mistral-releases-open-weights-for-mistral-large-4 ai|expert es 2026-10-30T15:04:14Z Mistral Large 4 iguala a GLM-5.3 en codificación con un 47% menos de tokens