Osmo claims its olfactory AI reaches human level, experts question it
Understand this story
AI startup Osmo claims its smell foundation model has achieved "human-level olfactory intelligence": approaching that of an average person in detecting and distinguishing odors, and approaching that of a junior perfumer in creating scents. The model was trained on more than 10 million human annotations and 800 million "odor tokens". Osmo says the model has already been used in the fragrance industry, and last month outperformed in-house perfumers at a major global consumer goods company in a comparison; it also screened 8 substances with better insect-repellent effects than DEET for the Gates Foundation, and partnered with StockX to identify counterfeit sneakers by smell. All of these results are Osmo's own claims. Experts have questioned the "human-level" wording.
AI overview based on source reports · Updated 23 hr ago
Coverage timeline
Follow the coverage to explore different perspectives.
- MIT Technology Review · AIThis smell startup says its AI nose is as good as a human’s
Osmo, an AI startup spun out of Google Brain in 2023, claims its foundation model for smell has reached "human-level olfactory intelligence"—matching the average person at detecting and discriminating odors and a junior perfume designer at creating them—and says its technology is already used in the fragrance industry. The company built a dataset of more than 10 million human labels and 800 million "scent tokens" using spectrometers and around 70 trained data labelers, and its smell-printing robot can design scents from a ChatGPT-style prompt. Smell scientist Avery Gilbert has criticized some of Osmo's marketing claims as undocumented, while former consultant Michael Schmuker says context is crucial to smell and sensor technology will need to improve for applications like food spoilage or medical diagnostics.
Activity for this story
- Current comparable activity
- 3
- Peak comparable activity
- 10Nov 18 20:00
- Change in the past 24 hours
- -50%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.