Osmo has achieved a remarkable reduction in artificial intelligence costs by 200 times following its transition to Meta Llama on Amazon Web Services (AWS). This strategic move enhances the company's ability to digitize scent and provides full control over its models and infrastructure, an important advancement in fragrance formula generation.
Creating fragrance formulas from creative briefs is inherently complex. Unlike other AI applications, scent evaluation requires physical production and expert assessment of the resulting fragrance. Richard Whitcomb, a key figure at Osmo, noted, "Unlike most AI systems, we can't evaluate results instantly. We have to physically produce the fragrance and have experts smell it." This reliance on physical evaluation has historically limited the speed and adaptability of fragrance creation processes.
Previously, Osmo used proprietary frontier model APIs, which proved unsustainable due to high costs, limited customization, and a lack of ownership over the models. The shift to Meta Llama has slashed training costs by a factor of 20 and revolutionized the company's approach to AI systems. Wesley Qian stated, "Moving to Llama on AWS changed how we think about building AI systems."
This transition aligns with Osmo's vision of treating fragrance creation similarly to text generation in language models. The process involves laying down major components of a fragrance first and then filling in the gaps—a methodology that resonates with how AI models learn patterns across diverse data sets. As Whitcomb explains, "We think about fragrance creation like writing—building formulas ingredient by ingredient, similar to how language models generate text."
The implications of this shift are significant. By adopting an open-model system, Osmo has matched the performance of established frontier models in evaluations conducted with trained perfumers. This not only validates its approach but also positions the company to innovate further within the fragrance industry.
The future looks promising for Osmo as it refines its AI capabilities in scent digitization. With significantly reduced costs and enhanced model ownership, the company is likely to explore new creative avenues in fragrance development, potentially reshaping how scents are designed and experienced. The integration of advanced AI technologies with traditional perfumery practices could lead to novel applications and products that appeal to a broader audience, marking a significant step forward in the intersection of technology and sensory experience.
