Atualizado diariamente3935 histórias verificadas
Notícias de IA. Sem o barulho.
Cobertura de IA verificada na fonte sobre lançamentos de produtos, mudanças políticas, pesquisas de segurança e movimentos do setor, explicada em inglês simples por uma equipe de educação sem fins lucrativos.
Fonte verificada
Cada história está vinculada às evidências mais fortes disponíveis: fontes originais, quando disponíveis, reportagens claramente atribuídas.
Inglês simples
O que aconteceu, por que é importante e o que assistir – sem jargão.
Sem preenchimento
Quando o sinal é fraco, não publicamos nada além de preencher o feed.
Mais histórias
9 históriasInovação
Preprint proposes a Bayesian way for LLM agents to recognize when they need stronger help
A new preprint studies agents that can transfer control to a stronger language model during reasoning, combining a Bayesian stopping rule with theoretical guarantees and a limited Qwen2.5-Coder validation.arxiv.orgProduto
Ars Technica reports IBM launches Granite 4.2 open-weight models for local AI
IBM’s Granite 4.2 family includes 3B, 8B and 30B open-weight models with a native 128,000-token context window. Ars Technica reports that the two larger versions received specialized training for tool use and agentic tasks.arstechnica.comInovação
PinSieve reports production gains from selective VLM serving and governed memory
A KDD 2026 workshop paper describes a deployed vision-language-model serving agent that handles unresolved content-quality cases, reporting higher review productivity, lower normalized operating cost and faster signal delivery.arxiv.orgEmpresa
loveholidays says Codex puts software building in reach of non-engineers
Online travel company loveholidays says Codex helped non-engineering teams build customer experiences and infrastructure changes while increasing AI-assisted code changes and deployments.
openai.comInovação
PuzzleKV proposes page-wise compression to reduce LLM KV-cache storage
A new arXiv paper proposes PuzzleKV, a training- and calibration-free method that independently compresses completed pages of large language model KV caches. It reports more than 96% of Full KV performance at about 60% of original storage, and more than 93% at 18.7% of original storage with quantization.arxiv.orgInovação
A bandit framework cuts LLM calls for automated essay-scoring prompt selection
A new arXiv paper reports that a multi-armed bandit controller selected prompting strategies for LLM essay scoring with comparable accuracy to exhaustive search while reducing calls by 78.4%.arxiv.orgPolítica
Bill Gates says AI has crossed multiple danger thresholds, MIT Technology Review reports
In an MIT Technology Review interview, Bill Gates says frontier AI has crossed thresholds involving biological misuse, cyberattacks, psychosocial dependence, job displacement and loss of control. He proposes monitoring models capable of creating novel molecules, international cooperation and new approaches to…technologyreview.comPolítica
South Korea deploys AI crypto-market surveillance as lawmakers seek wider FIU powers
CoinGeek reports that South Korea’s Financial Supervisory Service has deployed an AI system to flag suspicious crypto trading, while lawmakers propose giving the Financial Intelligence Unit direct investigative powers over unregistered virtual-asset businesses.coingeek.comIndústria
MiniMax revenue surges 283% but remains behind pace to meet forecast
South China Morning Post reports that Chinese AI company MiniMax increased first-half revenue 283% to US$116.6 million, driven by rapid enterprise growth, while losses remained substantial and revenue stayed below the pace needed to meet analysts’ full-year forecast.scmp.com
Um briefing útil por semana
Acompanhe a IA sem ficar no feed.
Receba as notícias verificadas sobre IA da semana, dados originais, ferramentas úteis, opções de aprendizado e novos trabalhos de IA.
Alcance pessoas que estão aprendendo IA
Contratar um profissional de IA ou lançar um produto útil de IA? Coloque isso na frente das pessoas que vieram aqui para aprender e agir.
Publique um trabalho de IAEnvie uma ferramenta de IA