Microsoft Muse, Helix and Google's AI Co-Scientist

AI’s taking the lab coat, the pen, and the canvas—meet Muse, Helix, and Co-Scientist, the brains behind the future!

February 23, 2025

Welcome back to Daily AI Skills.

Here’s what we are covering today:
1. Microsoft Muse - Ai Generated Gameplay and Controller Actions
2. Figure Helix - Humanoids in the Household
3. Google AI Co-Scientist - PhD level AI Model built with Gemini 2.0

Microsoft Unveils Muse: AI that can generate gameplay and controller action

Microsoft has introduced Muse, a groundbreaking generative AI model aimed at revolutionizing game development. Muse, developed in collaboration with Xbox Game Studios' Ninja Theory, is designed to enhance creativity and efficiency in game design.

Key Features of Muse:

  • Generative Capabilities: Muse can produce game visuals and predict controller actions, offering a dynamic tool for developers.

  • Training Data: The model was trained on over 1 billion images and actions from Ninja Theory's game Bleeding Edge, equating to more than seven years of continuous human gameplay.

  • Open Source Access: To foster further research and development, Microsoft has open-sourced Muse's weights and sample data, making them available through the WHAM Demonstrator on Azure AI Foundry.

Source: Microsoft

Helix: The Game-Changing AI That Lets Robots See, Think, and Act Like Never Before

This is a fascinating development in robotics! Helix launched by Figure seems to take a significant leap forward by integrating vision, language, and action into a single model that enables real-time generalization and dexterous control. Some standout aspects:

  • Real-time humanoid control: The full-upper-body control, including fine motor skills like finger movements, makes it well-suited for complex tasks.

  • Scalability through VLMs: Using Vision-Language Models (VLMs) to generalize skills instead of manually programming behaviours drastically improves adaptability.

  • Multi-robot collaboration: The ability to coordinate two robots to achieve a shared goal is rare in robotics.

  • Efficient onboard processing: Running on low-power GPUs means it’s commercially viable right away.

It feels like this could bridge the gap between robotic perception and real-world utility, especially for home automation and industrial logistics.

AI Co-Scientist: Accelerating Scientific Discovery with Gemini 2.0

AI Co-Scientist is a multi-agent AI system powered by Gemini 2.0, designed to assist researchers in generating novel hypotheses, refining research proposals, and accelerating breakthroughs in science and biomedicine.

Key Features:

  • Multi-Agent System – A coalition of specialized agents (Generation, Reflection, Ranking, Evolution, Proximity, Meta-review) works together to iteratively refine research hypotheses, mirroring the scientific method.

  • Automated Hypothesis Generation – AI synthesizes vast scientific literature, integrates cross-disciplinary insights, and generates novel, research-backed hypotheses tailored to specific goals.

  • Test-Time Compute Scaling – Self-play-based debates, ranking tournaments, and recursive critique ensure progressively improved outputs with higher scientific rigour.

  • Expert Collaboration – Researchers can seed ideas, provide real-time feedback in natural language, and guide AI-generated hypotheses towards practical applications.

  • Real-World Validation – Successfully identified and lab-validated drug repurposing candidates for acute myeloid leukemia (AML), showcasing its potential for biomedical breakthroughs.

Source: Google

Whether it’s painting, decoding DNA, or cooking up the next Nobel-worthy hypothesis, AI is getting alarmingly good at, well… everything. But don’t worry, humans still have the upper hand (for now).

Stay tuned as we keep up with the robots—before they start keeping up with us.

Share it with people you know are keeping pace with the changing AI world!