Shake‑up (Aug 3‑9): Nobel Laureate Leaves DeepMind, Jeff Dean Starts Startup
From Alphabet’s leadership reshuffle—Nobel laureate Demis Hassabis stepping down as DeepMind CEO and Jeff Dean launching a new AI venture—to Baidu’s Wenxin topping benchmarks, OpenAI’s security hire, SpaceX’s soaring AI‑driven revenue, MiniMax H3’s open‑source video model, FFmpeg 9.0’s release, and a novel AI‑text detection framework, this week’s roundup captures the most pivotal tech industry developments.
Alphabet experienced a major personnel shift: Nobel Prize winner Demis Hassabis stepped down as DeepMind CEO to become the company’s chief scientist and chair of DeepMind, focusing on long‑term AGI strategy and health‑science research via Isomorphic Labs, while Gemini 4 progress was noted. Senior staff Koray Kavukcuoglu took over day‑to‑day DeepMind management.
Simultaneously, Jeff Dean, Google’s chief scientist for 27 years, left to co‑found the public‑interest company Discovery Loop with three former Google engineers. Their mission is to automate the full research cycle—hypothesis generation, experiment execution, and result evaluation—to accelerate scientific discovery.
In other headline news, Baidu’s Wenxin Assistant achieved top scores on the SuperCLUE XClaw benchmark (97.62) and PinchBench v2 (94.6%), excelling in memory, data processing, content creation, code, and research analysis, while remaining free to users.
OpenAI announced the hiring of veteran security researcher Thomas Dullien (aka “the grandfather of hacking”), aiming to strengthen the company’s AI safety and network security capabilities amid rising concerns over model misuse.
SpaceX’s first post‑IPO financial report showed a 92 % YoY revenue surge, driven largely by AI services that contributed roughly a third of total revenue. However, capital expenditures surged, with most funds allocated to AI infrastructure, leading to volatile stock movements.
MiniMax released the open‑source video model H3 (code‑named “MiniMax H3”), delivering 2K resolution, dual‑channel audio, and advanced multimodal editing features. While performance is strong, deployment challenges include complex environment setup, high hardware requirements, and long rendering times.
To lower entry barriers, the platform Mysterious AI integrated MiniMax H3, offering web‑based video generation at low per‑second costs (0.009 CNY for 768p, 0.015 CNY for 2K) and an open API for developers.
FFmpeg version 9.0, codenamed “Lei,” was released to honor Chinese audio‑video pioneer Lei Xiaohua. The update adds Vulkan hardware acceleration, broader codec support, HDR processing, and native deep‑learning inference back‑ends.
At ACL 2026, researchers unveiled the RACE framework for fine‑grained AI‑generated text detection, distinguishing between original human text, AI‑generated text, AI‑assisted edits, and AI‑refined content by analyzing rhetorical structure trees with graph‑convolutional networks. Experiments show superior performance over existing baselines, though commercial deployment remains limited.
Additional industry notes covered MiniMax’s pricing drop, the rapid growth of flexible robotic “dexterous hands” with multi‑hundred‑million‑dollar funding, and product‑management insights from Zhang Fan on the evolving nature of AI‑centric product design.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
ZhongAn Tech Team
China's first online insurer. Through tech innovation we make insurance simpler, warmer, and more valuable. Powered by technology, we support 50 billion RMB of policies and serve 600 million users with smart, personalized solutions. ZhongAn's hardcore tech and article shares are here.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
