https://www.seventnews.com/en/articles/frances-ai-action-summit-put-governance-on-the-global-table SevenTnewS en 2026-07-29T16:00:00+02:00 France's AI action summit put governance on the global table https://www.seventnews.com/en/articles/frances-grand-ai-ambition-collides-with-government-inertia SevenTnewS en 2026-07-29T16:00:00+02:00 France's grand AI ambition collides with government inertia https://www.seventnews.com/en/articles/frances-25-billion-ai-push-81-labs-1000-startups-and-a-third-phase SevenTnewS en 2026-07-29T16:00:00+02:00 France's €2.5 billion AI push: 81 labs, 1,000 startups, and a third phase https://www.seventnews.com/en/articles/elevenlabs-just-turned-video-production-into-a-chat-conversation SevenTnewS en 2026-07-29T15:00:00+02:00 ElevenLabs just turned video production into a conversation https://www.seventnews.com/en/articles/microsofts-in-house-ai-models-are-cutting-costs-by-89-in-production SevenTnewS en 2026-07-29T14:45:58.932634+02:00 Microsoft's in-house AI models are cutting costs by 89% in production https://www.seventnews.com/en/articles/openai-says-its-health-ai-reasons-better-than-doctors-its-own-team-isnt-so-sure SevenTnewS en 2026-07-29T13:00:00+02:00 OpenAI's health AI claims to outthink doctors. Then its own lead walked it back. https://www.seventnews.com/en/articles/before-you-ask-an-ai-to-think-try-cleaning-its-input-first SevenTnewS en 2026-07-29T11:00:00+02:00 Clean your AI's input, not its reasoning. The speed gains are absurd. https://www.seventnews.com/en/articles/gsm8k-was-supposed-to-test-grade-school-math-up-to-42-of-its-questions-dont SevenTnewS en 2026-07-29T11:00:00+02:00 GSM8K Was Supposed to Test Grade-School Math. Up to 42% of Its Questions Don't https://www.seventnews.com/en/articles/why-the-best-graph-fraud-detector-on-three-benchmarks-doesnt-use-its-own-labels SevenTnewS en 2026-07-29T11:00:00+02:00 The fraud detector that got better by ignoring what it was told https://www.seventnews.com/en/articles/what-gemma-4-actually-knows-about-physics-a-new-paper-finds-the-representations-are-real SevenTnewS en 2026-07-29T11:00:00+02:00 Gemma 4 knows physics well enough to steer it, and that changes everything https://www.seventnews.com/en/articles/why-your-ais-chain-of-thought-is-wasting-tokens-and-how-optimal-stopping-fixes-it SevenTnewS en 2026-07-29T11:00:00+02:00 The hardest lesson for AI reasoning engines: when to shut up https://www.seventnews.com/en/articles/the-claude-code-playbook-is-now-a-community-wiki-83-tips-and-counting SevenTnewS en 2026-07-29T09:00:00+02:00 83 Claude Code tips just became a community wiki, and it reveals the new skill that matters https://www.seventnews.com/en/articles/distributional-rl-heads-make-risk-claims-that-are-mostly-false-a-new-audit-finds SevenTnewS en 2026-07-29T09:00:00+02:00 Distributional RL heads make risk claims that are mostly false. A new audit proves it. https://www.seventnews.com/en/articles/qwen3-2507-arrives-with-a-thinking-model-and-an-instruct-model-split-back-in-vogue SevenTnewS en 2026-07-29T09:00:00+02:00 Alibaba splits its flagship AI model in two because one size fits nobody https://www.seventnews.com/en/articles/apples-ios-in-brazil-compliance-not-capitulation SevenTnewS en 2026-07-29T08:00:00+02:00 Apple let Brazil inside its walled garden. The gates cost extra. https://www.seventnews.com/en/articles/microsoft-matches-gpt-56-in-excel-with-a-cheaper-older-gpu-model SevenTnewS en 2026-07-29T05:10:40.695552+02:00 Microsoft matches GPT-5.6 in Excel with a cheaper, older-gpu model https://www.seventnews.com/en/articles/microsofts-mai-cyber-1-flash-delivers-top-cybergym-scores-at-half-the-cost SevenTnewS en 2026-07-29T04:08:18.272825+02:00 Microsoft's MAI-Cyber-1-Flash delivers top CyberGym scores at half the cost https://www.seventnews.com/en/articles/why-the-best-general-ai-model-keeps-losing-to-narrower-ones-on-security SevenTnewS en 2026-07-28T21:00:00+02:00 Why the Best General AI Model Keeps Losing to Narrower Ones on Security https://www.seventnews.com/en/articles/the-model-race-has-a-second-track-now-orchestration-beats-raw-size SevenTnewS en 2026-07-28T19:00:00+02:00 The Model Race Has a Second Track Now: Orchestration Beats Raw Size https://www.seventnews.com/en/articles/openai-issues-minor-gpt-update-targeting-long-context-handling SevenTnewS en 2026-07-28T17:00:00+02:00 OpenAI shipped a long context fix. It just won't tell you how much better it got. https://www.seventnews.com/en/articles/frances-360m-bet-on-ai-nine-clusters-to-train-the-next-generation SevenTnewS en 2026-07-28T16:00:00+02:00 France's €360M bet on AI: nine clusters to train the next generation https://www.seventnews.com/en/articles/elevenlabs-signs-with-the-uk-government-to-bring-voice-ai-to-public-services SevenTnewS en 2026-07-28T15:00:00+02:00 ElevenLabs just got the UK government to put a testbed for voice AI in public services https://www.seventnews.com/en/articles/new-black-box-attack-reveals-vision-language-navigation-agents-are-easily-derailed-by-lens-like-noise SevenTnewS en 2026-07-28T15:00:00+02:00 A lens smudge just broke the AI that follows you through buildings https://www.seventnews.com/en/articles/the-fcc-is-about-to-ban-disguised-dji-gadgets-and-its-testing-new-retroactive-powers SevenTnewS en 2026-07-28T13:00:00+02:00 The FCC is about to retroactively ban DJI front companies, and that changes the rules for every imported gadget https://www.seventnews.com/en/articles/mmlu-ruled-ai-benchmarking-for-years-then-models-started-acing-the-answer-key SevenTnewS en 2026-07-28T11:00:00+02:00 MMLU Ruled AI Benchmarking for Years. Then Models Started Acing the Answer Key https://www.seventnews.com/en/articles/nvidias-av-flamingo-takes-on-long-video-understanding-where-most-models-fall-apart SevenTnewS en 2026-07-28T10:40:00+02:00 Nvidia's av-flamingo solves the one thing every other video AI gets wrong: time https://www.seventnews.com/en/articles/nvidias-4-billion-parameter-edge-model-does-something-most-small-vlms-dont SevenTnewS en 2026-07-28T10:39:00+02:00 Nvidia's 4B model does what most small robots can't: act, not just talk https://www.seventnews.com/en/articles/the-human-decisions-that-llms-can-replicate-better-than-surveys SevenTnewS en 2026-07-28T09:00:00+02:00 LLMs just bypassed the hardest part of behavioral economics: expensive surveys https://www.seventnews.com/en/articles/the-clapper-was-bad-on-purpose-a-tech-history-lesson SevenTnewS en 2026-07-28T09:00:00+02:00 The gadget that barely worked and changed everything anyway https://www.seventnews.com/en/articles/teslas-revenue-is-up-but-its-real-bet-is-spending-itself-into-the-red SevenTnewS en 2026-07-28T09:00:00+02:00 Tesla's revenue is up. Its real bet is spending itself into the red. https://www.seventnews.com/en/articles/stability-ai-counted-13-csam-reports-to-ncmec-in-its-first-transparency-report SevenTnewS en 2026-07-28T08:00:00+02:00 Stability AI's first transparency report counted 13 CSAM cases. The real story is what it doesn't say. https://www.seventnews.com/en/articles/frances-25b-ai-plan-is-more-than-a-budget-line-its-a-blueprint-for-national-ai-sovereignty SevenTnewS en 2026-07-27T23:57:04.343900+02:00 France's €2.5B AI plan: more than a budget, a blueprint for sovereignty https://www.seventnews.com/en/articles/ignore-the-new-models-claude-codes-real-update-is-in-the-margins SevenTnewS en 2026-07-27T22:31:04.001149+02:00 Ignore the new models. Claude Code's real update is in the margins https://www.seventnews.com/en/articles/the-748b-model-that-beats-gpt-55-and-claude-opus-48-by-splitting-itself-into-four-specialists SevenTnewS en 2026-07-27T21:03:05.537957+02:00 Macaron-V1-Venti's 748B four-specialist design beats GPT-5.5 and Opus 4.8 https://www.seventnews.com/en/articles/explainable-rl-takes-on-air-traffic-control-one-saliency-map-at-a-time SevenTnewS en 2026-07-27T18:17:27.570876+02:00 Explainable RL takes on air traffic control, one saliency map at a time https://www.seventnews.com/en/articles/four-chinese-ai-labs-just-rewrote-the-competitive-landscape-and-they-did-it-with-benchmarks-not-discounts SevenTnewS en 2026-07-27T18:10:00+02:00 Benchmarks, not bargains: China's AI labs reach parity https://www.seventnews.com/en/articles/the-claude-vs-fugu-calculus-when-a-swarm-beats-a-single-model SevenTnewS en 2026-07-27T18:09:00+02:00 The Claude vs Fugu calculus: when a swarm beats a single model https://www.seventnews.com/en/articles/kimi-k3-is-the-biggest-open-model-ever-its-still-not-the-best SevenTnewS en 2026-07-27T18:08:50.487262+02:00 Kimi K3 is the biggest open model ever. It's still not the best.