ChatGPT Learns Electromagnetic Physics
LLMs predicting metamaterial properties from simple text descriptions
Researchers demonstrate a fine-tuned LLM capable of predicting the physical properties of electromagnetic metamaterials from text descriptions alone.
- LLM trained on 40,000 data points accurately predicts absorptivity spectrum from text prompts describing metasurface geometry
- Outperforms conventional ML approaches with 10x less training data
- Demonstrates potential for LLMs to enable intuitive, natural language interfaces in complex engineering domains
- Creates new possibilities for engineering design through conversational AI
This research opens pathways for engineers to explore complex electromagnetic materials through simple text prompts, potentially accelerating design workflows and making advanced material science more accessible.