#022 The Limits of Embeddings, Out-of-Domain Data, Long Context, Finetuning (and How We're Fixing It)
How AI Is Built
00:00
Evolving Data, Evolving Models
Continuously adapting AI models to reflect changing enterprise data poses significant challenges, especially in terms of retaining context and evolving meanings. While manual interventions can align AI understanding with specific organizational terminology, automation remains elusive. The cycle of fine-tuning models with new data, such as the introduction of new systems, demands not only the recreation of extensive training datasets but also the indexing of new embeddings. Achieving seamless and ongoing fine-tuning for dynamic contexts could potentially streamline these processes, ensuring that AI systems remain relevant as organizational knowledge evolves.
Transcript
Play full episode
