Large language models can map the complex relationships between different smells even though they have never physically encountered a single scent.
April 24, 2026
Original Paper
Odor Maps from the LLM-derived similarity scores
arXiv · 2604.20310
The Takeaway
Artificial intelligence trained only on text can accurately calculate the similarity between odors like rose and lemon. These models produce odor maps that align almost perfectly with actual human sensory data and chemical evaluations. It was previously assumed that a sense as biological as smell required a physical nose and real-world experience to understand. The data shows that the way we describe smells in language contains enough hidden structure to recreate the entire olfactory system. This discovery opens the door for AI to design new fragrances or flavors using nothing but mathematical patterns in text.
From the abstract
The application of large language models (LLMs) to OdorSpace analysis attracts growing interest. Recent studies have explored the comparison of sensory evaluation spaces derived from LLMs with odor character profiles in the Dravnieks' dataset. In this study, we calculated pairwise distances of odor descriptors using three distance measures and statistically compared these LLM-derived similarities with distances derived from the original data. Next, we extended this approach to odor names (ingred