Good morning, AI enthusiasts!
AI is moving into interactive worlds, business software, and the research lab. Today’s lead is Genie 4, followed by Oracle’s expanded OpenAI partnership and a fresh wave of models built for specific jobs.
In today’s Mammoth:
THE QUICK SCOOP
WORLD MODELS
Summary: Google DeepMind introduced Genie 4, a real-time world model that generates scenes together with ambient audio, music, and dialogue. Users can steer unfolding events with natural-language instructions.
Quick Scoop:
The research preview is rolling out to trusted testers; broader access is planned, not generally available yet.
Gemini models help interpret goals and guide multi-step events inside the simulation.
Google says generated video and audio carry SynthID watermarks; object permanence, physics, and long-session consistency remain limitations.
Your AI Edge: For creative teams, this points toward faster interactive prototypes and scene planning. Treat simulations as experimental previews, not verified representations of how the real world behaves.
ENTERPRISE AI
Summary: Oracle and OpenAI expanded their partnership to connect Oracle applications and data with ChatGPT, while powering new Oracle offerings with OpenAI models and coding tools.
Quick Scoop:
Oracle AI Workspace is being developed to combine information and approved business workflows in ChatGPT.
App Composer is designed to build and deploy apps connected to approved Oracle data.
U.S. customers can purchase OpenAI API, Codex, and ChatGPT Work access through Oracle Marketplace using eligible Oracle Universal Credits.
Your AI Edge: Map one workflow that crosses finance, procurement, or planning systems. Check integration availability and permissions before assuming an announced product is ready for your organization.
SPONSORED BY MAMMOTHCLUB
Build the skills behind the tools with The Complete Python, Machine Learning, Bots and AI Mega Bundle+. Explore practical projects across data, machine learning, and AI applications.
Work with Python, Pandas, and NumPy.
Explore machine learning and deep learning workflows.
Build LLM applications with Python APIs and related frameworks.
AI IN SCIENCE
Summary: Anthropic disclosed its acquisition of Mithril Medicine and research partnerships with six drug-development organizations. The work explores how Claude can assist scientific research, laboratory coordination, manufacturing documentation, and trial planning.
Quick Scoop:
Mithril’s founder used Claude to coordinate operations and experiments across 11 external labs.
Anthropic reports promising mouse results for a muscular-dystrophy candidate, reached in less than 18 months.
Most reported work remains preclinical. Human trials for Mithril candidates could begin as early as late 2028 if further results hold up.
Your AI Edge: The practical opportunity is reducing bottlenecks across a research workflow. Early laboratory results do not establish that a treatment will be safe or effective in people.
SMALL MODELS
Summary: Liquid AI launched LFM3, an open-weight model family spanning text, vision, audio, omni, and decision tasks. The first release focuses on roughly two-billion-parameter models built on a shared backbone.
Quick Scoop:
Six models start from LFM3-1.7B-Base and can be adapted for specialized tasks.
Liquid says the text model fits in about 1 GB at four-bit precision and supports a 256K context window.
Weights, a playground, and API access are available, with support for local deployment and fine-tuning.
Your AI Edge: Test a narrow task on your actual target device. Measure memory use, response time, and accuracy under realistic context lengths before committing to a local deployment.
CREATIVE AI
Summary: Microsoft AI introduced MAI Image 2.7 in preview, focusing on image editing: relighting, restyling, restoring assets, and changing text or symbols while preserving the surrounding scene.
Quick Scoop:
Microsoft reports improvements over version 2.6 across the editing categories evaluated by Artificial Analysis.
The preview can be tried through MAI Playground, Arena, and Artificial Analysis.
The launch starts with editing; text-to-image generation is described as coming soon.
Your AI Edge: Try a difficult edit from your own workflow, such as moving a product into different lighting. Inspect shadows, lettering, and unchanged areas as closely as the requested modification.
SPONSORED BY MAMMOTHCLUB
Choose a recurring task and learn to connect its steps with Automate & Profit: The Complete Make.com Course Bundle+. Build practical workflows that put AI and automation to work.
Connect business and productivity tasks without writing code.
Explore chatbot and virtual-assistant workflows.
Apply prompt engineering within Make.com automations.
AGENT OPERATIONS
Summary: IBM’s TechXchange announcements bring agent identities, credential controls, discovery, and evaluation together across enterprise platforms. The emphasis is on knowing which agents exist, who owns them, and what they can do.
Quick Scoop:
Vault credential controls support limited access and short-lived credentials for agents.
watsonx Orchestrate discovery and isolated containment are in technology preview, including a shutdown control.
Evaluation capabilities are generally available, with quality checks and visibility into model and token usage.
Your AI Edge: Give every production agent an owner, a defined job, and access that can be revoked. Track its real actions and costs alongside output quality.
FORECASTING RESEARCH
Summary: Mistral’s new Timestral technical report introduces four time-series foundation models for probabilistic forecasting, with released weights and inference code.
Quick Scoop:
The models range from 5 million to 209 million parameters.
Their design supports information known about the future, such as scheduled events, alongside historical observations.
The authors report leading results on GIFT-Eval, TIME, and fev-bench; those benchmark results still need testing against your own data.
Your AI Edge: Compare forecasts with a simple baseline on held-out demand or operational data. Assess uncertainty estimates as well as average error, and guard against future information leaking into the test.
THE MAMMOTH LOWDOWN
⚙️ TOOLS FOR AI EDGE
Five tools to turn your data and AI ideas into usable applications:
Streamlit: Turn Python scripts and AI prototypes into interactive data apps.
Panel: Connect Python models, plots, and widgets in interactive dashboards.
Shiny for Python: Build reactive data and AI applications with Python.
Reflex: Build web applications with a Python framework and AI-assisted development tools.
Taipy: Create Python data and business-intelligence applications with customizable interfaces.
What Else Is Buzzing In AI
Claude’s government discount expands: Anthropic announced a 40% government discount covering Claude Enterprise, Claude for Government, and the API, with eligibility extending to state, local, tribal, and territorial governments.
Check Point shares a Daybreak case study: In a newly published customer report, Check Point says Daybreak reduced parts of its vulnerability-research workflow from hours to minutes.
Motley Fool introduces Fool 5Y: The company’s new AI-powered research score combines three proprietary methodologies to help premium members compare companies. It is a research aid, not a guaranteed return.