https://shift-ia.com/en/artigos/protostar-reusable-science/Shiften2026-11-08T12:17:21.857ZProtoStar puts scientific data reuse to a practical testhttps://shift-ia.com/artigos/protostar-ciencia-reutilizavel/Shiftpt2026-11-08T12:17:21.857ZProtoStar quer tornar pesquisa aberta utilizável por IAhttps://shift-ia.com/en/artigos/mistral-large4/Shiften2026-11-08T12:21:49.778ZMistral Large 4: check reasoning settings and licensing before integrationhttps://shift-ia.com/artigos/mistral-large4/Shiftpt2026-11-08T12:21:49.778ZMistral Large 4: confira licença e raciocínio antes de integrarhttps://shift-ia.com/en/artigos/typed-swift-decisions/Shiften2026-11-08T13:18:43.281884+00:00AnyDecisionModel gives Swift applications typed answers instead of prose to parsehttps://shift-ia.com/artigos/decisoes-tipadas-swift/Shiftpt2026-11-08T13:18:43.281884+00:00AnyDecisionModel troca a interpretação de prosa por respostas tipadas em Swifthttps://shift-ia.com/en/artigos/inkling-weight-release/Shiften2026-11-08T13:18:43.281884+00:00Inkling 2 Small reaches Tinker with its weight release still conditionalhttps://shift-ia.com/artigos/inkling-pesos-pendentes/Shiftpt2026-11-08T13:18:43.281884+00:00Inkling 2 Small chega ao Tinker; abertura dos pesos ainda depende de testeshttps://shift-ia.com/en/artigos/lfm3-modality-choice/Shiften2026-11-08T13:18:43.281884+00:00LFM3 makes modality selection part of model selectionhttps://shift-ia.com/artigos/lfm3-escolha-modalidade/Shiftpt2026-11-08T13:18:43.281884+00:00LFM3 organiza a especialização por modalidade, não por um modelo únicohttps://shift-ia.com/en/artigos/thinkingbox-database-outcomes/Shiften2026-11-08T13:18:43.281884+00:00ThinkingBox rewards agents for the database state they leave behindhttps://shift-ia.com/artigos/thinkingbox-resultado-no-banco/Shiftpt2026-11-08T13:18:43.281884+00:00ThinkingBox transforma o estado final do banco em recompensa para agenteshttps://shift-ia.com/en/artigos/trainium-support-lookup/Shiften2026-11-08T13:18:43.281884+00:00Check Trainium model support before provisioning a machinehttps://shift-ia.com/artigos/trainium-consulta-compatibilidade/Shiftpt2026-11-08T13:18:43.281884+00:00Consulte o suporte ao Trainium antes de provisionar a máquinahttps://shift-ia.com/en/artigos/anylanguagemodel-provider-choice/Shiften2026-11-08T14:01:47.078830+00:00AnyLanguageModel gives Swift apps a shared model interfacehttps://shift-ia.com/artigos/anylanguagemodel-escolha-do-provedor/Shiftpt2026-11-08T14:01:47.078830+00:00AnyLanguageModel reúne modelos locais e hospedados no Swifthttps://shift-ia.com/en/artigos/apertus-evaluation-records/Shiften2026-11-08T15:06:05.024191+00:00Apertus 1.5 makes evaluation conditions part of the releasehttps://shift-ia.com/artigos/apertus-avaliacao-reproduzivel/Shiftpt2026-11-08T15:06:05.024191+00:00Apertus 1.5 abre também o caminho até a avaliaçãohttps://shift-ia.com/en/artigos/autosynthdata-executable-training-tasks/Shiften2026-11-08T18:07:26.980723+00:00AutoSynthData makes the verifier part of the training examplehttps://shift-ia.com/artigos/autosynthdata-falhas-em-tarefas/Shiftpt2026-11-08T18:07:26.980723+00:00AutoSynthData transforma falhas em tarefas verificáveishttps://shift-ia.com/en/artigos/adapters-that-learn-the-harness/Shiften2026-11-08T12:15:01.105ZWhat if the model learned its harness?https://shift-ia.com/en/artigos/agent-memory-facts-procedures-evidence/Shiften2026-11-08T12:15:01.105ZAgent memory needs facts, procedures and evidencehttps://shift-ia.com/en/artigos/asr-accents-beyond-the-average/Shiften2026-11-08T12:15:01.105ZSpeech recognition needs an accent breakdownhttps://shift-ia.com/en/artigos/claude-code-skill-for-evidence-based-reading/Shiften2026-11-08T12:15:01.105ZBuild a Claude Code skill for verifiable readinghttps://shift-ia.com/en/artigos/claude-hooks-before-and-after-tools/Shiften2026-11-08T12:15:01.105ZClaude Code hooks: timing changes the controlhttps://shift-ia.com/en/artigos/claude-subagents-separate-context-not-usage/Shiften2026-11-08T12:15:01.105ZSubagents separate context, not usage limitshttps://shift-ia.com/en/artigos/cosmos-control-meets-synthetic-video/Shiften2026-11-08T12:15:01.105ZCosmos brings simulation controls into generated videohttps://shift-ia.com/en/artigos/cuga-hermes-why-the-harness-matters/Shiften2026-11-08T12:15:01.105ZCUGA and Hermes: the model is only part of the agenthttps://shift-ia.com/en/artigos/cwe-bench-repository-security-repairs/Shiften2026-11-08T12:15:01.105ZCWE-bench tests security repairs in repository contexthttps://shift-ia.com/en/artigos/dharma-extraction-beyond-transcription/Shiften2026-11-08T12:15:01.105ZDharma targets structured fields beyond OCRhttps://shift-ia.com/en/artigos/granite-switch-adapter-focused-patch/Shiften2026-11-08T12:15:01.105ZGranite Switch updates adapters while keeping the basehttps://shift-ia.com/en/artigos/hermes-gateway-browser-workspace/Shiften2026-11-08T12:15:01.105ZHermes brings its workspace to a browser through the gatewayhttps://shift-ia.com/en/artigos/lerobot-streaming-and-live-instructions/Shiften2026-11-08T12:15:01.105ZLeRobot links training data with live instructionshttps://shift-ia.com/en/artigos/mcp-setup-and-trust-boundaries/Shiften2026-11-08T12:15:01.105ZConnect MCP tools without losing track of trusthttps://shift-ia.com/en/artigos/palette-slide-design-workflow/Shiften2026-11-08T12:15:01.105ZPalette separates slide planning from drawinghttps://shift-ia.com/en/artigos/phonon-on-device-tts-and-licensing/Shiften2026-11-08T12:15:01.105ZPhonon puts speech on device—with a licensing choicehttps://shift-ia.com/en/artigos/quantization-model-hardware-and-task/Shiften2026-11-08T12:15:01.105ZChoose quantization for the model and the machinehttps://shift-ia.com/en/artigos/smolagents-code-as-an-action/Shiften2026-11-08T12:15:01.105Zsmolagents makes code an action—and execution a choicehttps://shift-ia.com/en/artigos/toyblocks-controlled-robot-learning/Shiften2026-11-08T12:15:01.105ZToyBlocks makes robot learning easier to inspecthttps://shift-ia.com/en/artigos/transformers-mic-small-realistic-checks/Shiften2026-11-08T12:15:01.105ZTransformers MIC shrinks checkpoints without discarding the testhttps://shift-ia.com/en/artigos/falcon-asr-dialect-evidence/Shiften2026-11-08T18:07:26.980723+00:00Falcon ASR's language support needs language-specific evidencehttps://shift-ia.com/artigos/falcon-asr-recorte-dialetal/Shiftpt2026-11-08T18:07:26.980723+00:00Falcon ASR: o recorte dialetal importa mais que a médiahttps://shift-ia.com/en/artigos/flux-region-control/Shiften2026-11-08T15:06:05.024191+00:00FLUX.3 lets prompts specify where objects belonghttps://shift-ia.com/artigos/flux-controle-por-regioes/Shiftpt2026-11-08T15:06:05.024191+00:00FLUX.3 transforma posição em parte do prompthttps://shift-ia.com/en/artigos/lfm-local-tool-execution/Shiften2026-11-08T15:06:05.024191+00:00LFM2.5 on-device: deciding an action is only the handoffhttps://shift-ia.com/artigos/lfm-agente-local-ferramentas/Shiftpt2026-11-08T15:06:05.024191+00:00LFM2.5 local: o modelo decide, o agente executahttps://shift-ia.com/en/artigos/ml-intern-training-acceptance/Shiften2026-11-08T18:07:26.980723+00:00ML Intern needs an acceptance criterion before a training runhttps://shift-ia.com/artigos/ml-intern-treino-com-criterio/Shiftpt2026-11-08T18:07:26.980723+00:00ML Intern: delegue o treino, preserve o critério de aceitehttps://shift-ia.com/en/artigos/muse-spark-video-sampling/Shiften2026-11-08T15:06:05.024191+00:00Muse Spark Open: video input starts with frame selectionhttps://shift-ia.com/artigos/muse-spark-video-amostragem/Shiftpt2026-11-08T15:06:05.024191+00:00Muse Spark Open: a amostragem decide o que o modelo vêhttps://shift-ia.com/en/artigos/tts-metrics-and-listening/Shiften2026-11-08T18:07:26.980723+00:00Open TTS Leaderboard measures different parts of speech qualityhttps://shift-ia.com/artigos/tts-metricas-e-escuta/Shiftpt2026-11-08T18:07:26.980723+00:00Open TTS Leaderboard separa inteligibilidade, voz e esperahttps://shift-ia.com/en/artigos/provenanceguard-claim-attribution/Shiften2026-11-08T18:07:26.980723+00:00ProvenanceGuard checks whether the citation belongs to the claimhttps://shift-ia.com/artigos/provenanceguard-origem-da-afirmacao/Shiftpt2026-11-08T18:07:26.980723+00:00ProvenanceGuard verifica de onde veio cada afirmaçãohttps://shift-ia.com/artigos/adapters-modelo-aprende-harness/Shiftpt2026-11-08T12:15:01.105ZE se o modelo aprendesse o harness que vai usar?https://shift-ia.com/artigos/asr-sotaques-fora-da-media/Shiftpt2026-11-08T12:15:01.105ZReconhecimento de fala precisa ouvir quem fica fora da médiahttps://shift-ia.com/artigos/claude-code-skill-leitura-verificavel/Shiftpt2026-11-08T12:15:01.105ZComo criar uma skill de leitura no Claude Codehttps://shift-ia.com/artigos/claude-hooks-antes-depois-da-acao/Shiftpt2026-11-08T12:15:01.105ZHooks no Claude Code: o momento da ação muda o controlehttps://shift-ia.com/artigos/claude-subagentes-contexto-nao-gratis/Shiftpt2026-11-08T12:15:01.105ZSubagentes separam contexto, não eliminam consumohttps://shift-ia.com/artigos/cosmos-simulacao-controle-e-imagem/Shiftpt2026-11-08T12:15:01.105ZCosmos aproxima controle de simulação e geração visualhttps://shift-ia.com/artigos/cuga-hermes-peso-do-harness/Shiftpt2026-11-08T12:15:01.105ZCUGA e Hermes: o modelo não explica o agente inteirohttps://shift-ia.com/artigos/cwe-bench-correcao-seguranca-repos/Shiftpt2026-11-08T12:15:01.105ZCWE-bench coloca a correção de segurança dentro do repositóriohttps://shift-ia.com/artigos/dharma-extrair-campos-nao-so-transcrever/Shiftpt2026-11-08T12:15:01.105ZDharma mira campos estruturados, não apenas texto reconhecidohttps://shift-ia.com/artigos/granite-switch-patch-nos-adapters/Shiftpt2026-11-08T12:15:01.105ZGranite Switch atualiza adapters sem trocar o modelo-basehttps://shift-ia.com/artigos/hermes-gateway-desktop-no-navegador/Shiftpt2026-11-08T12:15:01.105ZHermes leva o workspace ao navegador pelo próprio gatewayhttps://shift-ia.com/artigos/lerobot-dados-instrucoes-e-execucao/Shiftpt2026-11-08T12:15:01.105ZLeRobot aproxima dados de treinamento e instruções em execuçãohttps://shift-ia.com/artigos/mcp-conectar-sem-confundir-confianca/Shiftpt2026-11-08T12:15:01.105ZMCP na prática: conectar ferramentas sem apagar fronteiras de confiançahttps://shift-ia.com/artigos/memoria-agente-fatos-procedimentos-evidencia/Shiftpt2026-11-08T12:15:01.105ZMemória de agente: fatos, procedimentos e evidência não são a mesma coisahttps://shift-ia.com/artigos/palette-slides-com-linguagem-visual/Shiftpt2026-11-08T12:15:01.105ZPalette trata slides como planejamento e desenhohttps://shift-ia.com/artigos/phonon-voz-local-licencas/Shiftpt2026-11-08T12:15:01.105ZPhonon leva voz ao dispositivo, mas as licenças variamhttps://shift-ia.com/artigos/quantizacao-memoria-hardware-e-tarefa/Shiftpt2026-11-08T12:15:01.105ZQuantização: escolha pelo modelo e pelo hardwarehttps://shift-ia.com/artigos/smolagents-codigo-como-acao/Shiftpt2026-11-08T12:15:01.105Zsmolagents usa código como ação — e exige escolher o executorhttps://shift-ia.com/artigos/toyblocks-mundos-controlados/Shiftpt2026-11-08T12:15:01.105ZToyBlocks quer tornar o aprendizado robótico mais examinávelhttps://shift-ia.com/artigos/transformers-mic-verificacoes-pequenas-reais/Shiftpt2026-11-08T12:15:01.105ZTransformers MIC: modelos pequenos para erros que passam em silênciohttps://shift-ia.com/en/artigos/pytorch-mps-metal-kernels/Shiften2026-11-08T14:01:47.078830+00:00PyTorch MPS changes kernels—and fixes image outputhttps://shift-ia.com/artigos/pytorch-mps-kernels-metal/Shiftpt2026-11-08T14:01:47.078830+00:00PyTorch no Mac: MPS muda kernels e corrige saídashttps://shift-ia.com/en/artigos/beam-memory-beyond-active-experts/Shiften2026-11-08T15:06:05.024191+00:00Beam: sparse computation still needs room for the weightshttps://shift-ia.com/artigos/beam-memoria-alem-dos-experts/Shiftpt2026-11-08T15:06:05.024191+00:00Beam: poucos parâmetros ativos não significam pouca memóriahttps://shift-ia.com/en/artigos/torchneuron-operation-trace/Shiften2026-11-08T14:01:47.078830+00:00TorchNeuron: trace an operation all the way to hardwarehttps://shift-ia.com/artigos/torchneuron-rastro-da-operacao/Shiftpt2026-11-08T14:01:47.078830+00:00TorchNeuron: siga a operação até o hardwarehttps://shift-ia.com/en/artigos/torchtpu-native-device/Shiften2026-11-08T14:01:47.078830+00:00TorchTPU brings a native device path to PyTorchhttps://shift-ia.com/artigos/torchtpu-dispositivo-nativo/Shiftpt2026-11-08T14:01:47.078830+00:00TorchTPU aproxima a TPU do código PyTorch existentehttps://shift-ia.com/en/artigos/vocabx-vocabulary-cost/Shiften2026-11-08T14:01:47.078830+00:00VocabX separates cheaper output from compressed decisionshttps://shift-ia.com/artigos/vocabx-custo-do-vocabulario/Shiftpt2026-11-08T14:01:47.078830+00:00VocabX: comprimir a resposta sem comprimir a decisão