In the context of LLMs, graphs or their simpler cousins - trees are surprisingly good as both inputs and outputs.
My team builds LLM-based features as part of our b2b SaaS product.
The primary challenge is how to get fuzzy, incomplete, heterogeneous business knowledge into LLMs and do something useful with it. Trees have been excellent. Even simple cases like summarizing content, there is a big improvement in relevance, accuracy and precision.
My team builds LLM-based features as part of our b2b SaaS product.
The primary challenge is how to get fuzzy, incomplete, heterogeneous business knowledge into LLMs and do something useful with it. Trees have been excellent. Even simple cases like summarizing content, there is a big improvement in relevance, accuracy and precision.