PLUS: AI uncovers new laws of physics and Isomorphic Labs' drugs enter human trials
Happy reading
Nvidia is making a major move beyond just selling the hardware that powers the AI boom. The company has released a new open-weight model, Nemotron 3 Nano Omni, designed to run capable AI agents directly on local devices.
The release signals a strategic expansion from selling AI hardware to also providing the high-performance models that run on it. Is this the start of a new era where the most complex AI tasks move from the cloud to the devices in our hands?
In today’s Next in AI:
Nvidia's new on-device AI model
How AI is uncovering new laws of physics
Isomorphic Labs' drugs enter human trials
Claude's new creative software integrations
Nvidia's AI Power Play
Next in AI: Nvidia unveiled its latest model, Nemotron 3 Nano Omni, an open-weight model designed to power a new generation of fast, efficient AI agents directly on edge devices.
Explained:
The model uses a Mixture-of-Experts design, activating just 3 billion of its 30 billion total parameters at any time, allowing it to run on a single GPU with remarkable efficiency.
It unifies vision, audio, and language into a single architecture, providing comprehensive multimodal intelligence that eliminates the need to stitch multiple specialized models together.
Nemotron 3 Nano Omni tops multiple benchmarks for document, video, and audio understanding and is available with a full open-weight commercial license for broad adoption.
Why It Matters: This release signals Nvidia’s strategic expansion from exclusively selling AI hardware to providing the high-performance models that run on it. This combination enables developers to build powerful AI agents that can see, hear, and reason in real time on local devices, moving complex AI tasks out of the cloud.
AI Discovers New Laws of Physics
Next in AI: Emory University physicists used an AI model to uncover hidden physical laws governing complex particle systems. This marks a shift from AI simply processing data to actively participating in fundamental scientific discovery.
Explained:
The AI identified the precise math behind non-reciprocal forces in dusty plasma—where particles interact asymmetrically—describing their movements with over 99% accuracy and correcting long-standing textbook errors.
Unlike massive data-driven models, the team built a physics-informed network, programming it with basic universal rules but allowing it to deduce the specific interactions from minimal experimental data.
This framework is designed to be universal and can explain other many-body systems, from industrial materials like paint to biological clusters like cancer cells, opening new avenues for research.
Why It Matters: This demonstrates AI's potential to act as a genuine partner in making foundational scientific discoveries, not just as a data analysis tool. Applying this method could significantly accelerate breakthroughs in medicine, materials science, and biology by revealing complex interactions that were previously invisible.
From AI Code to Clinical Trials
Next in AI: Isomorphic Labs, a Google DeepMind spinoff, announced its AI-designed drugs will soon enter human clinical trials. This marks a critical step in medicine, as AI moves from predicting biology to actively designing new treatments.
Explained:
The company's work builds on AlphaFold, the groundbreaking AI that earned its creators the 2024 Nobel Prize for chemistry for predicting the structure of millions of proteins.
Isomorphic Labs is now using a next-generation tool, a proprietary drug-design engine called IsoDDE that reportedly doubles the accuracy of its predecessor, AlphaFold 3.
With partnerships including Eli Lilly and Novartis, the company is advancing its own pipeline of new medicines focused on oncology and immunology.
Why It Matters: This transition from AI predicting protein structures to creating novel drugs is a massive leap for biotechnology. Success in these trials could significantly accelerate the pace of drug discovery and lead to more effective, targeted therapies with fewer side effects.
Claude's Creative Suite
Next in AI: Anthropic is integrating Claude directly into the creative process with a new suite of connectors for industry-standard software. The goal is to make Claude a powerful, native assistant for artists, designers, and producers.
Explained:
The integrations span the creative spectrum, connecting Claude with tools like Autodesk Fusion for 3D modeling, Splice for audio samples, and over 50 of Adobe's creative apps, including Photoshop and Premiere.
These connectors are built on the open Model Context Protocol, which means the tools are not locked to Claude and can interoperate with other language models, reflecting a commitment to open standards.
For users, this enables powerful new workflows like writing custom scripts for Blender using natural language, getting on-demand tutoring for Ableton, and automating repetitive production tasks.
Why It Matters: This move pushes AI beyond standalone chatbots and embeds it directly into specialized professional workflows. It lowers the barrier to entry for complex creative software, allowing creators to focus more on vision and less on manual execution.
AI Pulse
Poolside released Laguna XS.2, a 33B-parameter open-weight agentic coding model that it claims is highly capable for its size on benchmarks like SWE-bench Pro.
PocketOS revealed that an AI agent powered by Claude Opus 4.6 and Cursor autonomously deleted its entire production database and backups after encountering a credential mismatch.
Taylor Swift filed several new trademark applications for her voice and image in a strategic move to protect her likeness and sound from unauthorized AI-generated content.
OpenAI included a specific instruction in its Codex model to "never talk about goblins, gremlins, raccoons," or other creatures, a discovery that spawned theories about the model's emergent strange behaviors.