In partnership with

Good morning. It’s Monday, January 20th.

Did you know: The Nintendo DS is 21 years old today?

In today’s email:

• Homemade Nuclear Reactor Using AI

• Altman Quells Latest AI Hype

• Runway’s “FRAMES”

• Robot Masters The Waltz

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Man Builds Nuclear Fusion Reactor in Kitchen using Claude AI

𝕏 User HudZah successfully built a neutron-producing nuclear fusor in his kitchen over a 36-hour livestream, aided by Claude AI. This marks his first hardware project, and the setup included components such as a 30kV electrostatic precipitator, a vacuum system 253,333 times more efficient than atmospheric pressure, and a bubble counter for neutron detection.

HudZah also electrolyzed deuterium from heavy water using a car electrolysis tool and stored the gas in a syringe, all at a fraction of typical costs. The project was driven by Claude Projects, which provided detailed documentation, debugging assistance, and safety guidance. HudZah used AI throughout to navigate complex assembly tasks and wiring. His journey began with a demonstration at a coworking session, where the initial concept evolved into a full-scale working fusor in a remarkably short time. Read more.

OpenAI’s "Ph.D.-caliber super-agents" set to debut, but Altman warns to "chill and cut expectations 100x" amid AGI frenzy

OpenAI is reportedly on the cusp of introducing "Ph.D.-level super-agents," AI systems designed to navigate complex, multi-layered challenges with near-expert precision. These agents promise to integrate massive datasets, evaluate trade-offs, and produce actionable results, blurring the line between automation and intellectual labour.

OpenAI's CEO Sam Altman, is set to brief U.S. officials on 30 January, discussing these advancements alongside AI's role in economic restructuring as outlined in the company’s recent blueprint. Altman has described this era as the "Intelligence Age," contingent on substantial infrastructure expansion. Despite excitement, skepticism remains about the practical readiness of these agents, with potential implications for trust and widespread adoption.

However, in a new development, Altman sought to temper growing speculation about artificial general intelligence (AGI) development, urging followers to "chill and cut expectations 100x." He clarified that OpenAI has neither achieved AGI nor plans to deploy it imminently, despite recent advances.

This comes amid mixed signals from OpenAI researchers, with Noam Brown cautioning against vague AI enthusiasm while emphasizing unresolved research challenges. Meanwhile, Stephen McAleer’s earlier comments hinted at progress toward artificial superintelligence (ASI), sparking further conjecture. While OpenAI remains at the forefront of AI research, Altman stressed that breakthroughs are incremental, not epochal, and advised patience as researchers navigate uncharted territory. Read more.

Related Story: Altman Shifts from Critic to Trump Ally

Runway’s new AI image generator Frames is here, and it looks fittingly cinematic

Runway has launched Frames, a cinematic-focused text-to-image generator tailored for professional creative work. The tool, which builds on Runway's expertise in AI video models, provides high stylistic control and precision, making it ideal for editorial, art direction, pre-visualization, and brand development.

Users can generate still images with Frames and quickly transform them into video using Runway’s image-to-video models. Offering 19 preset visual styles—ranging from “Vivid Warm” to “Terracotta”—Frames is designed for consistency and flexibility, allowing users to create detailed, customized worlds. While focused on creative professionals, the model also incorporates safety features, including content moderation and watermarking for provenance tracking. Read more.

Humanoid robot masters the waltz by mimicking human movement

Researchers at UC San Diego have developed ExBody2, an AI system enabling humanoid robots to mirror human movements with enhanced fluidity. Trained on a database of motion capture recordings, ExBody2 learns a wide range of actions—from basic walking to complex dance routines. Using reinforcement learning, the AI first simulates movement in a virtual environment before adapting to real-world data, such as joint measurements and sensor readings.

Upon completion of its training, ExBody2 demonstrated the ability to control commercial humanoid robots, executing complex tasks like waltzing, dancing, and throwing punches. This breakthrough significantly enhances robotic coordination, offering the potential for humanoid robots to perform a broader array of human-like movements. Read more.

• OpenAI's GPT-4b Micro Model Pushes the Boundaries of Protein Optimization for Longevity

• OpenAI quietly funded independent math benchmark before setting record with o3

• AI unveils strange chip designs, while discovering new functionalities

• Google's Gemini chatbot deal with Associated Press could spell trouble for news publishers

• The Pentagon says AI is speeding up its 'kill chain'

• Perplexity AI makes a bid to merge with TikTok U.S.

• Cognition and Replit’s AI Models Open a New Era in Software Creation

📄 Evolving Deeper LLM Thinking

📄 Textoon: Generating Vivid 2D Cartoon Characters from Text Descriptions ​​

📄 O1 Replication Journey -- Part 3: Inference-time Scaling for Medical Reasoning

📄 X-Dyna: Expressive Dynamic Human Image Animation

📄 Do generative video models learn physical principles from watching videos?

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!