CLaim Offer: Sign-up for a Maintenace Plan Get a Free Website Redesign

May 15, 2025
Episode 334: AI Generated Digest – Continuous Thought Machines, Figma AI Features, LegoGPT
In this first AI-generated digest episode, we explore three significant AI developments: Sakana AI’s Continuous Thought Machine (CTM) architecture, Figma’s new AI features announced at Config 2025, and Carnegie Mellon University’s LegoGPT. This episode is narrated by an AI version of Alex created using an AI voice clone, with full disclosure to listeners about the […]

In this first AI-generated digest episode, we explore three significant AI developments: Sakana AI’s Continuous Thought Machine (CTM) architecture, Figma’s new AI features announced at Config 2025, and Carnegie Mellon University’s LegoGPT. This episode is narrated by an AI version of Alex created using an AI voice clone, with full disclosure to listeners about the AI-generated nature of the content. The host requests honest feedback from listeners about this experimental format, particularly for digest-style episodes covering AI news.

Keywords

  • AI Generated Content
  • Continuous Thought Machine
  • Sakana AI
  • Figma AI Features
  • LegoGPT
  • Neural Networks
  • AI Architecture
  • Voice Cloning
  • Figma Make
  • Figma Buzz
  • Text-to-Code
  • Image Generation
  • AI Website Builder
  • Lego Instructions
  • Neuron Timing
  • AI Disclosure
  • Design Tools
  • Marketing Automation
  • AI Benchmarks
  • Content Creation

Key Takeaways

Continuous Thought Machine (CTM)

  • New model architecture from Sakana AI mimicking brain’s neural networks
  • Embeds neuron timing into reasoning process unlike standard LLMs
  • Each artificial neuron has internal clock to synchronize thoughts
  • Model iterates in clock cycles, predicting next layer in each cycle
  • Takes multiple passes at tasks similar to human cognition
  • Processes images by “looking around” multiple times
  • Uses step-by-step incremental reasoning
  • Shows impressive results in unstructured reasoning tasks
  • Represents significant advancement in AI reasoning approaches
  • Differs from transformer architecture used in typical LLMs

Figma AI Features

  • Figma Make: AI prompt-to-code tool for turning static designs into interactive prototypes
  • Allows designers to use natural language instructions to animate elements
  • Figma Buzz: Tool specifically for marketing and brand teams to create on-brand graphics at scale
  • Library of templates with design interface for rapid content creation
  • Enhanced AI features in Figma Design and FigJam for advanced image generation and editing
  • Smart AI helpers for brainstorming content ideas
  • Figma Sites: New website builder with AI prompts to add functionality to static designs
  • Interactive elements generated through natural language instructions
  • Positions Figma to compete with Canva in marketing design space
  • Streamlines workflow between design and implementation

LegoGPT

  • AI model from Carnegie Mellon University Generative Intelligence Lab
  • Converts natural language text prompts into buildable Lego instructions
  • Creates step-by-step building sequences from simple descriptions
  • Will be open-released at some point in the future
  • Potential marketing applications for branded Lego constructions
  • Example: “small red house with a blue roof” generates corresponding instructions
  • Represents practical and fun AI use case to drive adoption
  • Demonstrates creative application of generative AI beyond standard use cases
  • Potentially valuable for educational and entertainment purposes
  • Shows AI’s ability to generate physical construction blueprints

Links

⁠https://www.therundown.ai/p/openai-goes-global-with-stargate⁠

⁠https://www.therundown.ai/p/ai-predicts-cancer-outcomes-from-selfies⁠

⁠https://sakana.ai/ctm/#:~:text=Sakana%20AI%20is%20proud%20to,toward%20bridging%20the%20gap%20between⁠

⁠https://ai-rockstars.com/continuous-thought-machine-how-sakana-ai-integrates-biological-thinking-into-ai-systems/#:~:text=,and%20increases%20performance%20and%20interpretability⁠

⁠https://arxiv.org/abs/2505.05522#:~:text=representation,versatility%20across%20a%20range%20of⁠

⁠https://www.anthropic.com/research/tracing-thoughts-language-model⁠

⁠https://openai.com/index/learning-to-reason-with-llms/⁠

⁠https://www.figma.com/blog/config-2025-press-release/⁠

⁠https://help.figma.com/hc/en-us/articles/30965205437975-What-s-new-from-Config-2025#:~:text=Requires%20a%20Full%20seat⁠

⁠https://www.infoq.com/news/2025/05/legogpt-text-prompts/#:~:text=Researchers%20at%20Carnegie%20Mellon%20University,manually%20or%20by%20robotic%20systems⁠

author avatar
Alex Carlson

Recent Episodes

Episode 331: GenSpark – The All Purpose AI Superagent

Episode 331: GenSpark – The All Purpose AI Superagent

In this episode, we explore GenSpark, a comprehensive AI super agent platform that offers a wide range of integrated tools and capabilities for marketers. Founded in 2023 by former Baidu employees Eric Jing and Kay Zhu, GenSpark uses a "mixture of agents" architecture...

read more

Let’s Get Started

Ready To Make a Real Change? Let’s Build this Thing Together!