In the dynamic landscape of artificial intelligence (AI), large language models (LLMs) have garnered immense attention and admiration for their ability to understand and generate human-like text.
However, it’s crucial to note that while LLMs can play a critical role for many organizations, they are not standalone solutions. Lily AI, a trailblazer in the field of purpose-built AI for retailers and brands, harnesses the potential of large language models (LLMs) to revolutionize retail experiences. Lily AI’s approach is unique in that it fuses LLM capabilities with human expertise in the fashion, home, and beauty domains, resulting in industry-specific AI solutions that are increasing sales for retailers, enhancing the shopping experience for consumers, and shaping the future of retail.
The Problem With Using ChatGPT for Product Synonyms
Large language models like ChatGPT have undeniably opened doors to numerous possibilities, but they are still a work in progress. Despite their capabilities, LLMs are in a constant state of refinement and learning.
At Lily AI, we recognize the value of human critical assessment in training these models. This is particularly evident in the realm of product synonyms, where the nuances and intricacies of industry-specific language require human domain expertise and oversight for accuracy. In contrast to a generic LLM approach, Lily AI’s models are meticulously trained and honed for the retail domain. Our experts work hand in hand with AI, collaborating with the machine learning teams to train the AI on the frontend and meticulously assess and fine-tune its output on the tail end. This coalescence of machine learning and human expertise ensures that the final output is not only accurate but also aligned with industry standards and consumer expectations.
Complexity of Synonyms: Unraveling the Language Maze
Synonyms, often viewed as simple word replacements, harbor a complexity that is often overlooked. To truly understand the challenges that AI faces in this domain, let’s delve into the intricacies of synonyms, hypernyms, hyponyms, and word variations. As we all know, synonyms are words that share similar meanings, yet too often, they are confused with hypernyms which encompass broader categories and hyponyms which represent narrower classifications. Word variations further complicate matters by adding a layer of diversity that requires nuanced comprehension. And if the incorporation of these words isn’t implemented correctly, there is a great risk of negatively affecting relevance and shopper experience, let alone achieving positive experience, which is always our ultimate goal.
ChatGPT vs. Lily AI
To illustrate the real-world application of these concepts, let’s review examples from the realms of fashion, home, and beauty. While ChatGPT’s attempts at generating synonyms may be commendable, it often falls short when it comes to capturing industry-specific jargon and nuances.
- Fashion: ChatGPT offered synonyms for “slip dress” such as “camisole dress,” “lingerie dress,” “nightgown dress,” “sheath dress,” and “bodycon dress,” but these miss the mark when it comes to capturing what shoppers are actually looking for and the type of language they use when searching, which Lily AI’s domain experts can seamlessly identify, i.e., “silky dress.”
- Home: In the home decor space, ChatGPT struggled to distinguish between “sofa,” “divan,” and “settee,” whereas Lily AI’s purpose-built AI understands the subtle variations in customer preferences and uses, i.e., “couch.”
- Beauty: ChatGPT provided synonyms for “lipstick” like “lip color,” but it did not grasp the nuance of a synonym like “lip rouge” which is a crucial differentiator in the beauty industry.
In the example below, Lily AI domain experts audited ChatGPT’s results for ‘slip dress’ and ‘boho chic’ synonyms. In the ‘slip dress’ example, ChatGPT failed to provide any accepted synonyms. In the ‘boho chic’ example, ChatGPT provided some accepted results yet was missing other synonym variations. In both examples, the synonyms returned included were rejected for either being inaccurate altogether, or a hyponym or word variation.
Purpose-Built AI Pioneering Retail Transformation
The transformational potential of AI in retail lies not in horizontal, generic models, but in purpose-built solutions that cater to industry-specific demands.
Lily AI’s approach stands as a testament to this philosophy, showcasing the synergy between LLMs and human expertise. While LLMs provide excellent capabilities, the true magic unfolds when industry specialists collaborate to assess, refine and enhance the output. As the investment community increasingly recognizes the significance of purpose-built AI, Lily AI’s journey continues to illuminate the path toward retail innovation that seamlessly marries the power of technology with the finesse of human insight.
