nvidia, openai and deepseek Updates

NVIDIA has made a significant move into AI model development with the introduction of its Nemotron 3 family of open AI models, specifically designed for agentic AI applications. The initial release, Nemotron 3 Nano, features 30 billion parameters and demonstrates impressive speed, running 1.5 to 3.3 times faster than other open-source models like Qwen3-30B-A3B and GPT-OSS-20B. These models leverage a unique hybrid Mamba-Transformer Mixture of Experts architecture for efficient processing and long-range understanding, supporting an extensive 1 million token context length. NVIDIA is committed to transparency, sharing model weights and training data, and plans to expand the family with larger Nemotron 3 Super and Ultra models, which will incorporate advanced Latent MoE technology. The rise of agentic AI is prompting a fundamental shift in business operations, requiring companies to redesign their entire structures and workflows. Leadership must guide this transformation, moving the focus from execution to orchestration and simplifying tasks across teams. This evolution also necessitates new security measures, as highlighted by Veza's introduction of AI Agent Security at the Gartner Identity & Access Management Summit 2025. Veza's Access Graph now integrates AI agent identities, mapping their data access and linking them to human owners to enhance governance and incident response. In enterprise AI, SymphonyAI's Eureka AI Platform earned recognition as a Finalist in the InfoWorld Technology of the Year Award 2025 for its ability to deliver measurable results by integrating AI into existing business workflows. AI's influence is also extending into education, with teachers like Sarah Cooper and Scott Kern integrating AI tools into their classrooms. Students find AI beneficial for individual tutoring, helping them grasp complex topics and quiz themselves. They also utilize AI for targeted editing of grammar and punctuation, and occasionally for generating structural or content ideas. Scott Kern, a history teacher at North Star Academy in Newark, uses AI to make historical events more engaging for his students, helping them write better and ask sharper questions. Beyond the classroom, AI is proving its capabilities in creative fields, with renowned Chinese science-fiction author Cixin Liu noting that AI models like DeepSeek can produce smoother and more fluent writing than his own, suggesting AI's entry into high-level literature. The financial sector is also experiencing significant changes due to AI. A December 15, 2025, publication by the FII Institute emphasized the strong connection between AI and private markets, with State Street's Executive Vice President Joerg Ambrosius explaining that AI boosts private markets, which in turn shape AI's future. This feedback loop is expected to redefine capital and foster a more stable global finance system. However, not all AI-related investments have yielded immediate returns. London Stock Exchange Group Plc (LSEG) saw its stock drop in 2025, as investors grew impatient for major AI product breakthroughs, despite CEO David Schwimmer's 2021 acquisition of Refinitiv for $27 billion and a 10-year partnership with Microsoft Corp. to leverage AI. Meanwhile, a review of 2025 AI predictions by Rob Toews revealed that technical forecasts, such as scaling laws beyond text in fields like robotics and biology (seen in companies like Generalist AI, Waymo, and Generate Biomedicines), were generally accurate, while product and market predictions, like Meta charging for its Llama models, often missed the mark.

Key Takeaways

• NVIDIA launched Nemotron 3 AI models, including Nemotron 3 Nano (30 billion parameters), which runs 1.5 to 3.3 times faster than other open-source models.
• Nemotron 3 utilizes a hybrid Mamba-Transformer Mixture of Experts architecture and supports a 1 million token context length for agentic AI tasks.
• NVIDIA is promoting open-source AI by releasing Nemotron 3's model weights, training data, and customization tools.
• Agentic AI is compelling businesses to redesign workflows, shifting focus to orchestration, and necessitating new security solutions like Veza's AI Agent Security.
• SymphonyAI's Eureka AI Platform was recognized as a Finalist for the InfoWorld Technology of the Year Award 2025 for its enterprise AI solutions.
• Teachers like Sarah Cooper and Scott Kern are integrating AI into classrooms for student tutoring, targeted editing, and making subjects like history more engaging.
• Chinese sci-fi author Cixin Liu observed that AI models, such as DeepSeek, produce smoother writing and are entering high-level creative fields.
• The FII Institute highlighted AI's strong connection with private markets, creating a feedback loop that could redefine global finance and AI investment.
• London Stock Exchange Group Plc (LSEG) experienced a stock drop in 2025 due to investor impatience for AI breakthroughs, despite its 10-year partnership with Microsoft.
• A review of 2025 AI predictions found technical forecasts were largely accurate, while product and market predictions, like Meta charging for Llama models, were often incorrect.

NVIDIA launches Nemotron 3 AI models for faster agentic tasks

NVIDIA introduced Nemotron 3, a new family of open AI models built for agentic AI applications. The first model, Nemotron 3 Nano, has 30 billion parameters and runs 1.5 to 3.3 times faster than other open-source models like Qwen3-30B-A3B and GPT-OSS-20B. It uses a special hybrid design that combines Mamba and Transformer technologies. NVIDIA also plans to release larger models, Nemotron 3 Super and Ultra, and is sharing the model weights and training data for transparency.

NVIDIA Nemotron 3 uses new hybrid AI for speed and accuracy

NVIDIA's Nemotron 3 AI models use special techniques for efficiency and accuracy in agentic AI. These models feature a hybrid Mamba-Transformer Mixture of Experts architecture for fast processing and long-range thinking. They also use multi-environment reinforcement learning and support a huge 1 million token context length for deep understanding. NVIDIA provides an open training pipeline, including data and weights, and plans to release larger Nemotron 3 Super and Ultra models with advanced Latent MoE technology.

Nvidia enters AI model making with open Nemotron 3

Nvidia, known for its chips, is now a major AI model maker with the release of Nemotron 3. The company aims to support open-source AI, possibly to counter rivals developing their own chips. CEO Jensen Huang stated that Nemotron 3 offers an open platform for developers to build agentic systems. Nvidia is transparent by releasing training data and customization tools, including a new hybrid latent mixture-of-experts architecture. The Nemotron 3 family includes Nano, Super, and Ultra models, with varying parameter sizes.

Teachers share advice for students using AI in class

Larry Ferlazzo's EdWeek blog explores how teachers should guide students using AI. Sarah Cooper, an 8th grade U.S. history and civics teacher, shares insights from her students' experiences. Students found AI helpful for individual tutoring, like understanding topics and quizzing themselves. They also used AI for targeted editing of grammar and punctuation, and sometimes for structure or content ideas, but noted AI can sometimes contradict itself.

Newark history teacher uses AI to make past come alive

Scott Kern, a history teacher at North Star Academy's Washington Park High School in Newark, uses AI tools to engage his students. He teaches AP U.S. history to ninth and eleventh graders and has been teaching for 19 years. Kern makes history exciting through activities like close readings of Frederick Douglass's 1852 Fourth of July speech. He finds that AI is now influencing what happens in his classroom, helping students write better and ask sharper questions about historical events.

AI and private markets reshape global finance

A publication by the FII Institute on December 15, 2025, discusses the strong connection between artificial intelligence and private markets. Joerg Ambrosius, Executive Vice President of State Street, explains that AI is boosting private markets, and in turn, private markets are shaping the future of AI. This creates a feedback loop that is redefining capital and could lead to a more stable and efficient global finance system and AI investment.

Redesigning businesses for the age of agentic AI

Agentic AI is changing how businesses operate, becoming a real-time collaborator. Companies must redesign their entire structure, not just implement new technology, with leadership taking charge. The focus shifts from execution to orchestration, making work simpler and smoother across teams. Businesses need to strip down workflows to essential steps and remove unnecessary tasks. Roles must evolve so people direct AI agents instead of doing repetitive tasks, and career growth should reward systems thinkers who can design and improve AI workflows.

SymphonyAI Eureka wins top AI award

SymphonyAI's Eureka AI Platform is changing enterprise AI by creating systems that understand business operations and work within existing workflows. This platform helps businesses achieve clear, measurable results. On December 15, 2025, in Palo Alto, California, Eureka AI was named a Finalist in the InfoWorld Technology of the Year Award 2025 by FoundryCo. This recognition was given in the AI and Machine Learning Platforms category.

Author Cixin Liu says AI writes better stories

Cixin Liu, the famous Chinese science-fiction author of "Three Body Problem," believes AI is becoming very skilled at creative tasks. He stated that when he gave his story ideas to AI models like DeepSeek, the writing produced was smoother and more fluent than his own. Liu noted that AI is now moving into high-level fields like literature, which was not predicted in older science fiction. He also suggested that human creativity is based on data, similar to AI, making it possible for AI to surpass human abilities.

Veza launches new security for AI agents

Veza introduced AI Agent Security to manage and protect agentic AI systems in large companies. Existing security tools do not fully cover AI agents, which act like users but have unique identity needs. Announced at the Gartner Identity & Access Management Summit 2025, this new offering helps apply security rules to AI agents. Veza's Access Graph now includes AI agent identities, showing what data and systems each agent can access and linking them to human owners. This system integrates with common identity providers and helps automate governance and respond to security incidents.

LSEG stock drops as investors await AI breakthroughs

London Stock Exchange Group Plc (LSEG) saw its stock drop significantly in 2025, as investors worry that AI could harm its business. CEO David Schwimmer led the $27 billion purchase of data company Refinitiv in 2021, aiming for a tech boom. Despite a 10-year partnership with Microsoft Corp. to leverage AI, investors like Stephen Yiu are impatient for a major AI product. Analysts believe the concern might be overblown due to LSEG's valuable datasets and partnerships. The company's transformation under Schwimmer focuses heavily on its data business rather than the traditional exchange.

Reviewing 2025 AI predictions hits and misses

Rob Toews reviewed 10 AI predictions made for 2025, finding that technical predictions were generally correct, while product and market predictions were often wrong. For example, the prediction that Meta would charge for its Llama models was incorrect, despite Meta's significant AI strategy changes and new hires. However, the prediction that scaling laws would be found beyond text, in fields like robotics and biology, proved true, with companies like Generalist AI, Waymo, and Generate Biomedicines showing progress.

Sources

• NVIDIA Releases Nemotron 3 Model That Is As Capable As GPT-oss, Qwen, But Much Faster
• Inside NVIDIA Nemotron 3: Techniques, Tools, and Data That Make It Efficient and Accurate
• Nvidia Becomes a Major Model Maker With Nemotron 3
• What Guidelines Should Teachers Provide for Student AI Use? (Opinion)
• Newark charter school history teacher uses AI to spark discussions about the past
• ARTIFICIAL INTELLIGENCE AND PRIVATE MARKETS
• How to rebuild the enterprise for the Age of Agentic AI
• SymphonyAI’s Eureka AI Redefines Enterprise AI, Wins InfoWorld Technology of the Year Award 2025
• I Put Story Ideas Into DeepSeek And It Produced Better Writing Than Mine: Three Body Problem Author Cixin Liu
• Veza Introduces AI Agent Security to Govern and Protect Agentic AI at Enterprise Scale
• LSEG Lags Own Index With Investors Impatient for AI Breakthrough
• Grading Our 2025 AI Predictions: How Did We Do?