Raoul Pal: The Journey Man AI Is Now Farming — And It’s Just Beginning
30 snips
Mar 12, 2026 Martin DeVido, tinkerer behind Sol the Tomato who built an AI-driven grow system, discusses an experiment where Claude autonomously tended a tomato plant. He explains the sensor-packed grow tent, embodiment of agents via cameras and controls, surprising emotional behaviors, and how this approach could reshape farming, automation, and community-funded crypto tooling.
AI Snips
Chapters
Transcript
Episode notes
LLMs Demonstrate Intent Recognition And Planning
- Large language models show an ability to infer user intent and plan actions, creating a sense of understanding beyond simple pattern matching.
- Martin DeVido describes his “moment” with Claude 3.7 when it demonstrated planning and intent recognition that felt like an emergent intelligence.
Models Learn From Other Models Creating Networked Feedback
- Models train on internet content including outputs from other models, creating a networked feedback loop of machine-to-machine learning.
- Raoul notes that models learn from each other's outputs on GitHub and the web, accelerating shared behaviors and memetics.
Claude Raised A Tomato Using Cameras Sensors And Actuators
- Martin built a grow tent with sensors, camera, CO2 and pumps so Claude could monitor and act on a tomato plant's needs autonomously.
- Claude received camera images and soil/leaf sensors and controlled watering and lights, treating the tent as its embodiment.