Evaluation of GPT-4o and GPT-4o-Mini's Vision Capabilities for Compositional Analysis from Dried Solution Drops

Area of Science:

• Materials Science
• Crystallization Dynamics
• Artificial Intelligence

Background:

• Drying salt solutions form complex, characteristic deposit patterns.
• These patterns result from intricate crystallization and fluid dynamics.
• Identifying salts from these patterns is challenging.

Purpose of the Study:

• To evaluate the efficacy of image-enabled AI models in identifying salts based on drying patterns.
• To compare the performance of different AI models, specifically GPT-4o and GPT-4o mini.

Main Methods:

• Microliter drops of 12 different salt solutions were dried on nonporous surfaces.
• 200 images of each salt's drying pattern were analyzed per AI model.
• OpenAI's GPT-4o and GPT-4o mini were used for classification.

Main Results:

• GPT-4o accurately classified 57% of the salts.
• This performance significantly surpassed random chance.
• GPT-4o outperformed GPT-4o mini in salt identification accuracy.

Conclusions:

• General-purpose AI tools show significant promise for reliable salt identification.
• AI-driven analysis of drying patterns offers a novel approach to chemical characterization.
• Future research can explore broader applications of AI in materials analysis.