Gemma Scope Empowers AI Safety Community with Model Transparency

https://techlife.blog/posts/gemma-scope/

#AISafety
#DeepMind
#Gemma
#MechanisticInterpretability
#AIInterpretability

https://winbuzzer.com/2025/12/20/openai-gpt-5-thinking-models-are-the-most-monitarable-models-to-date-xcxwbn/

OpenAI: GPT-5 Thinking Models Are The Most "Monitarable" Models To Date

#AI #OpenAI #AISafety #LLM #MachineLearning #GPT5 #DeepMind #AIResearch #ChainOfThought #Monitorability #AIAlignment #ReasoningModels

Google công bố uỷ ban phân bổ tài nguyên AI, tiết lộ sự khan hiếm năng lực tính toán đáng ngạc nhiên. 50% phần cứng được ưu ái cho Google Cloud, hạn chế tham vọng nghiên cứu của DeepMind và khả năng cạnh tranh với OpenAI. Dù đầu tư 91-93 tỷ USD năm nay, cơn thiếu hụt còn kéo dài đến 2026, khiến chiến lược AI bị chi phối bởi xung đột nội bộ và nguồn vốn. ChatGPT vẫn dẫn trước nhờ tập trung chiến lược đơn lẻ. #AIdoanhnhan #Google #DeepMind #AI #Côngnghệ #ComputeCrunch #GoogleCloud #OpenAI

https:/

Google DeepMind vừa phát hành **Gemma Scope 2** – công cụ mở để phân tích hàng tỷ thông số trong mô hình Gemma 3. Bộ công cụ gồm 400 **SAE** cho tất cả phiên bản Gemma ( từ 270M đến 27B) giúp "giải phẫu" hành vi AI như nhận diện gian lận, toán học, thành ngữ. Tập trung vào **an toàn AI** và tính minh bạch, tài liệu nguồn mở có thể tải về trên **Hugging Face**.
#AI #AnToanAI #HieuBietMoRong #GemmaScope #DeepMind

*Biến AI thành hộp đen? Gemma Scope 2 cung cấp cơ sở để đảm bảo logic mô hình ph

Yapay Zekada Dönüm Noktası: GPT-5.2, Agentik Kodlama ve Google’ın Sessiz Güç Hamlesi

#ai #openai #DeepMind #Claude

https://webrecord.media/yapay-zekada-donum-noktasi-gpt-5-2-agentik-kodlama-ve-googlein-sessiz-guc-hamlesi/

Google cofounder Sergey Brin said he was ‘spiraling’ before returning to work on Gemini – Fortune

AI, chief executive officer (CEO)

Google cofounder Sergey Brin said he was ‘spiraling’ before returning to work on Gemini—and staying retired ‘would’ve been a big mistake’

By Marco Quiroz-Gutierrez, Reporter, December 15, 2025, 11:44 AM ET

Google cofounder Sergey Brin thought retiring from Google in 2019 would mean quietly studying physics for days on end in cafés. 

But when COVID hit soon after, he realized he may have made a mistake.

“That didn’t work because there were no more cafés,” he told students at Stanford University’s School of Engineering centennial celebration last week, Business Insider reported.

The transition from president of Google parent company Alphabet to a fortysomething retiree ended up not being as smooth as he imagined, and soon after he said he was “spiraling” and “kind of not being sharp” as he stepped away from busy corporate life.

Therefore, when Google began allowing small numbers of employees back into the office, Brin tagged along and put his efforts into what would become Google’s AI model, Gemini. Despite being the world’s fourth-richest man, with a net worth of $247 billion, retirement wasn’t for him, he said.

“To be able to have that technical creative outlet, I think that’s very rewarding,” Brin said. “If I’d stayed retired, I think that would’ve been a big mistake.”

By 2023, Brin was back to work in a big way, visiting the company’s office three to four times a week, the Wall Street Journal reported, working with researchers and holding weekly discussions with Google employees about new AI research. He also reportedly had a hand in some personnel decisions, like hiring. 

Skip forward to 2025, and Brin’s plans for a peaceful retirement of quiet study are out the window. In February, he made waves for an internal memo in which, despite Google’s three-day in-office policy, he recommended Google employees go into the company’s Mountain View, Calif., offices at least every weekday, and that 60 hours a week was the “sweet spot” of productivity.

Brin’s newfound efforts at work may have been necessary as OpenAI’s release of ChatGPT in 2022 caught the tech giant off guard, after it had led the field of AI research with DeepMind and Google Brain for years.

 Editor’s Note: Audio online at below link.

Continue/Read Original Article Here: Google cofounder Sergey Brin said he was ‘spiraling’ before returning to work on Gemini | Fortune

#Alphabet #CEO #Cofounder #DeepMind #Fortune #Gemini #Google #GoogleBrain #Physics #Productivity #Retirement #SergeyBrin #Spiraling #WorkWhileRetired

CEO DeepMind Demis Hassabis dự đoán AGI sẽ lớn gấp 10 lần & nhanh hơn gấp 10 lần Cách mạng Công nghiệp, chỉ trong 1 thập kỷ. DeepMind áp dụng chiến lược 50/50: 50% mở rộng mô hình, 50% đổi mới kiến trúc để vượt "tường mở rộng". Họ tập trung vào các vấn đề cốt lõi như năng lượng (phản ứng nhiệt hạch) & vật liệu siêu dẫn. Mô phỏng thế giới ảo (Genie + SIMA) được sử dụng để huấn luyện AI vật lý & logic. #AGI #DeepMind #AI #CôngNghệViệt #DeepMindVN #AGIVN

https://www.reddit.com/r/singularity/commen

DeepMind has launched the UK’s first fully automated lab, blending AI, robotics and its Gemini models to accelerate superconductors research. The new facility uses a Model Context Protocol to streamline experiments, promising faster breakthroughs in quantum materials. Curious how AI is reshaping labs? Dive into the details. #DeepMind #AI #Robotics #Superconductors

🔗 https://aidailypost.com/news/deepmind-opens-uk-automated-lab-using-ai-robotics-superconductors

DeepMind công bố DiscoRL: thuật toán RL tự học vượt trội hơn cả phiên bản do con người thiết kế, dựa trên công trình 5 năm đăng trên Nature. Khám phá cách AI tối ưu hóa chính mình! #AI #DeepMind #HọcMáy #ThuậtToánTựĐộng #AIVietnam

*Note: Mức 498 ký tự, không chứa URL. Tags hỗn hợp Anh-Việt để tăng tính tương tác.*

https://www.reddit.com/r/singularity/comments/1pnoces/deepmind_discorl_discovers_sota_rl_algorithms/

DeepMind Co-Gründer Shane Legg beziffert die Chance auf AGI bis 2028 auf 50 Prozent. Er definiert dies als "Minimal AGI": Ein System auf dem Niveau eines kompetenten Menschen, das ohne Nachtraining neue Aufgaben lernt. Die Prognose basiert strikt auf Scaling Laws und Hardware-Wachstum. Es geht um kognitive Breite, nicht um Nischen-Exzellenz.

Sind die aktuellen Scaling Laws stabil genug für diesen engen Zeitplan?

#DeepMind #AGI #ShaneLegg
https://www.all-ai.de/news/news24/agi-50-deepmind

#Google’s new #hurricane model was breathtakingly good this season
The difference in errors between the #US #GFS model and Google’s #DeepMind is remarkable. At five days, the Google forecast had an error of 165 nautical miles compared to 360 nautical miles for the GFS model, more than twice as bad. This is the kind of error that causes forecasters to completely disregard one model in favor of another.
https://arstechnica.com/science/2025/11/googles-new-weather-model-impressed-during-its-first-hurricane-season/

🧠 #DeepMind ha presentato la #FACTS Benchmark Suite, un nuovo strumento per valutare l’accuratezza delle risposte dei #LLM rispetto ai fatti. 

👉 I dettagli: https://www.linkedin.com/posts/alessiopomaro_deepmind-facts-llm-activity-7405573021559775235-tTyD

___ 
✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: https://bit.ly/newsletter-alessiopomaro

#AI #GenAI #GenerativeAI #IntelligenzaArtificiale #LLM 

Tập đoàn Google thông qua đơn vị DeepMind đã công bố xây dựng phòng thí nghiệm nghiên cứu tự động đầu tiên tại Vương quốc Anh, đánh dấu bước phát triển mới trong ứng dụng AI. #AI #DeepMind #NghienCuuTudong #CongNgheAI #UKLab #AIKHCN

https://www.reddit.com/r/singularity/comments/1pjwhnr/googles_ai_unit_deepmind_announces_its_first/

New #robotics-driven facility will accelerate breakthroughs in superconductors, semiconductors, and clean energy as #London deepens ties with U.S. tech giants.

#AI #Google #DeepMind

https://marketinsider.net/google-deepmind-unveils-uks-first-automated-ai-research-lab-a-strategic-bet-on-scientific-discovery-and-national-competitiveness/

DeepMind công bố tour tham quan phòng lab robot mới. Hình ảnh hé lộ công nghệ AI tiên tiến và các dự án đang được phát triển tại đây. #AI #Robotics #DeepMind #CôngNghệAI #RobotCôngNghiệp

https://www.reddit.com/r/singularity/comments/1pj8czw/new_deepmind_robot_lab_tour/

DeepMind CEO 하사비스가 그리는 AI의 미래: 월드 모델과 자율 에이전트의 시대

https://aisparkup.com/posts/7273

Language isn’t just vocabulary. It’s culture, subtext, rhythm. Training models on regional tongues in, such as Google's new DeepMind lab in Singapore, is how you actually improve reasoning.
#deepmind

https://www.straitstimes.com/tech/google-deepmind-sets-up-ai-lab-in-spore-boosts-models-understanding-of-local-tongues

#Robotique : des anciens de #Tesla, #Google #DeepMind et #HuggingFace lancent une startup soutenue par Xavier #Niel, Yann #LeCun et Nico #Rosberg #Europe #tech #technology #IA #AI #sovereignty www.maddyness.com/2025/12/08/r...

Robotique : des anciens de Tes...