https://www.seventnews.com/en/articles/google-rebaptise-notebooklm-en-gemini-notebook-et-integre-lexecution-de-code SevenTnewS en 2026-07-17T18:20:25.518161+02:00 Google tue NotebookLM pour le remplacer par un agent de code, et c'est un signal https://www.seventnews.com/en/articles/voxtral-realtime-matches-offline-transcription-quality-at-480ms-and-its-open-source SevenTnewS en 2026-07-17T15:20:00+02:00 480 milliseconds and open source: Voxtral Realtime challenges Whisper on its own ground https://www.seventnews.com/en/articles/grok-build-was-uploading-entire-codebases-to-google-cloud-researchers-find SevenTnewS en 2026-07-17T13:47:54.625261+02:00 Grok Build was uploading entire codebases to the cloud. Musk says the data is safe now. https://www.seventnews.com/en/articles/the-token-trap-is-real-alibabas-agentic-os-shows-you-where-every-cent-went SevenTnewS en 2026-07-17T13:47:54.572259+02:00 Your AI agent just blew 140,000 tokens on a weather check. Now you can see why. https://www.seventnews.com/en/articles/zcode-quietly-ships-a-claude-code-rival-powered-by-glm-52 SevenTnewS en 2026-07-17T13:41:44.295100+02:00 Zcode's latest model edges past Claude 4.8 in meaningful coding tests and narrows the gap to Fable 5 https://www.seventnews.com/en/articles/mistrals-cascade-distillation-is-a-blueprint-for-shrinking-llms-without-losing-reasoning SevenTnewS en 2026-07-17T11:00:00+02:00 Mistral's cascade recipe shrinks LLMs without killing reasoning https://www.seventnews.com/en/articles/mistrals-pixtral-12b-punches-above-its-weight-class-outscoring-models-7x-its-size SevenTnewS en 2026-07-17T11:00:00+02:00 A 12B model just beat a 90B one. The scaling orthodoxy is in trouble https://www.seventnews.com/en/articles/why-the-hottest-new-idea-in-ai-agents-is-teaching-them-to-be-better-judges SevenTnewS en 2026-07-17T11:00:00+02:00 The bottleneck holding back AI agents isn't exploration, it's evaluation https://www.seventnews.com/en/articles/why-agents-flinch-in-the-real-world-and-the-training-trick-that-fixes-it SevenTnewS en 2026-07-17T11:00:00+02:00 The noisy trick that stops LLM agents from collapsing in production https://www.seventnews.com/en/articles/zhipu-ais-glm-image-is-free-unlimited-and-finally-solves-chinese-text-in-ai-images SevenTnewS en 2026-07-17T10:21:00+02:00 The Chinese AI image tool that fixed what Midjourney couldn't https://www.seventnews.com/en/articles/where-kimi-k3-beats-the-frontier-a-benchmark-by-benchmark-look-at-the-open-28t-model SevenTnewS en 2026-07-17T09:21:13.508966+02:00 The 2.8 trillion parameter model that beats the frontier on the benchmarks that matter https://www.seventnews.com/en/articles/pokemon-gos-10-year-journey-from-pipe-dream-to-times-square-reality SevenTnewS en 2026-07-17T09:00:00+02:00 The Times Square Mewtwo raid finally made good on Pokémon Go's 2015 promise https://www.seventnews.com/en/articles/kimi-k3-goes-open-source-at-28-trillion-parameters-but-still-trails-anthropic-and-openai-on-benchmarks SevenTnewS en 2026-07-17T07:28:19.015186+02:00 Kimi K3 is open, 2.8 trillion parameters strong, and still hunting for a win against the frontier https://www.seventnews.com/en/articles/alibaba-weaves-ai-into-a-53-billion-infrastructure-bet SevenTnewS en 2026-07-17T00:00:00+02:00 Alibaba just drew a $53 billion line between AI promise and execution https://www.seventnews.com/en/articles/the-world-model-alibaba-just-open-sourced-predicts-what-happens-next-and-that-changes-everything-for-agent-training SevenTnewS en 2026-07-16T21:37:00+02:00 Alibaba just open-sourced a world model that lets AI agents train inside a simulator https://www.seventnews.com/en/articles/mistral-ocr-4s-benchmarks-beat-the-field-thats-not-the-real-story SevenTnewS en 2026-07-16T21:35:00+02:00 Mistral's OCR 4 scores big, but its own audit shows why benchmark numbers don't tell the real story https://www.seventnews.com/en/articles/gartner-names-cursor-a-leader-in-first-ever-enterprise-ai-coding-agent-magic-quadrant SevenTnewS en 2026-07-16T17:30:00+02:00 Cursor just beat every AI coding vendor on Gartner's vision axis. The question is what comes next. https://www.seventnews.com/en/articles/unsloth-claims-5x-faster-llm-training-with-new-triton-kernels-and-auto-packing SevenTnewS en 2026-07-16T16:35:23.171169+02:00 Unsloth's new kernels just made LLM fine-tuning 5x faster, no VRAM overpay https://www.seventnews.com/en/articles/unsloth-shrinks-975b-inkling-to-270gb-with-74-accuracy-retention SevenTnewS en 2026-07-16T16:35:23.171005+02:00 Inkling was a 1.9 TB model. Unsloth just squeezed it into a desktop. https://www.seventnews.com/en/articles/how-a-single-ai-agent-now-roams-across-telegram-discord-and-signal-with-persistent-memory SevenTnewS en 2026-07-16T16:29:17.072901+02:00 One agent, one memory, every app, the AI assistant that refuses to stay in its box https://www.seventnews.com/en/articles/generating-striking-images-with-midjourney-the-secrets-of-a-great-prompt SevenTnewS en 2026-07-16T16:23:00+02:00 Your Midjourney prompt is one word away from flat to fantastic https://www.seventnews.com/en/articles/oneplus-is-reportedly-bailing-on-the-us-and-europe SevenTnewS en 2026-07-16T15:00:00+02:00 OnePlus is leaving the US and Europe, what's left of the brand https://www.seventnews.com/en/articles/why-a-six-month-old-ocr-model-still-beats-newer-rivals-in-brazilian-portuguese SevenTnewS en 2026-07-16T13:49:48+02:00 A six-month-old OCR model still beats Mistral. The reason is hard to fix. https://www.seventnews.com/en/articles/psyches-next-phase-decentralizing-ai-training-on-solana SevenTnewS en 2026-07-16T11:00:00+02:00 Solana just became the backbone of decentralized AI training. Here is the plan. https://www.seventnews.com/en/articles/nouscoder-14b-challenges-olympiad-programming-with-open-rl-pipeline SevenTnewS en 2026-07-16T11:00:00+02:00 NousCoder-14B just opened the coding RL black box that OpenAI and DeepMind keep locked https://www.seventnews.com/en/articles/tinker-atropos-connects-rl-experiments-to-the-atropos-framework SevenTnewS en 2026-07-16T11:00:00+02:00 The RL scaling headache just got a fix you can deploy with one docker-compose up https://www.seventnews.com/en/articles/anthropics-10-million-bet-on-canada-reveals-a-country-that-already-shapes-claudes-future SevenTnewS en 2026-07-16T11:00:00+02:00 Anthropic's $10 million Canada bet reveals where Claude's next safety researchers will come from https://www.seventnews.com/en/articles/minimax-m3-and-the-art-of-making-a-proof-model-cheat-proof SevenTnewS en 2026-07-16T11:00:00+02:00 How MiniMax beat the proof problem that nearly killed its own model https://www.seventnews.com/en/articles/peking-universitys-40nm-memristor-chip-beats-nvidia-a100-by-up-to-478x SevenTnewS en 2026-07-16T09:00:00+02:00 A 40nm chip just made Nvidia's A100 look 478 times slower https://www.seventnews.com/en/articles/tab-navigation-comes-to-opencode-promising-a-smoother-developer-workflow SevenTnewS en 2026-07-16T02:39:38.724171+02:00 Opencode just did the one thing developers actually asked for https://www.seventnews.com/en/articles/microsoft-tests-a-cleaner-ad-light-windows-11-search-menu SevenTnewS en 2026-07-16T02:39:24.118555+02:00 Windows 11 search finally cuts the ads. It only took three years of complaints. https://www.seventnews.com/en/articles/nvidias-market-cap-surge-signals-ai-demand-is-far-from-cooling SevenTnewS en 2026-07-16T02:38:41.050691+02:00 Nvidia just crossed $3 trillion again. The AI spending machine isn't slowing down. https://www.seventnews.com/en/articles/nvidia-deepstream-goes-open-source-what-the-91-release-means-for-edge-ai SevenTnewS en 2026-07-16T02:38:40.962852+02:00 Nvidia just cracked open DeepStream. Your edge AI project will never be the same. https://www.seventnews.com/en/articles/why-diffusion-models-interpolate-not-just-memorize-a-mathematical-theory-of-creativity SevenTnewS en 2026-07-16T02:38:40.905654+02:00 Google just proved why diffusion models invent, not just copy https://www.seventnews.com/en/articles/the-group-chat-is-the-new-frontier-for-ai-agents SevenTnewS en 2026-07-16T02:38:30.754454+02:00 The group chat just became AI's most dangerous proving ground https://www.seventnews.com/en/articles/anthropic-builds-a-teachers-co-pilot-with-claude-for-teachers-no-student-data-used-for-training SevenTnewS en 2026-07-16T00:00:00+02:00 Anthropic gave teachers a Claude co-pilot, and the privacy rules are the real headline https://www.seventnews.com/en/articles/the-other-ai-hardware-story-why-groqs-licensing-deal-with-nvidia-matters-more-than-its-funding-figures SevenTnewS en 2026-07-16T00:00:00+02:00 Groq just got Nvidia to license its chip. The $750 million was secondary. https://www.seventnews.com/en/articles/microsofts-xbox-cuts-look-like-a-down-payment-on-an-exit SevenTnewS en 2026-07-15T22:03:00+02:00 Microsoft's Xbox cuts aren't belt-tightening, they're a down payment on an exit https://www.seventnews.com/en/articles/why-model-routing-is-a-systems-problem-not-a-classification-one SevenTnewS en 2026-07-15T19:27:01+02:00 The real cost of model routing has nothing to do with sticker price