Despite significant investments in training large language models, integrating them into practical applications remains challenging. CEO Jonas Andrulis of Aleph Alpha emphasizes that fine tuning is often overrated as a solution for model limitations. Though it can adjust style, it doesn't effectively impart new knowledge. An alternative, Retrieval Augmented Generation (RAG), allows models to access and update external information databases without retraining. However, its effectiveness hinges on the proper documentation of knowledge and relevant data alignment with the model’s training set, which is often a hurdle in many organizations.
"About a year ago, it felt that everybody was under the assumption that fine tuning is this magic bullet. The AI system doesn't do what you want it to do? It just has to be fine tuned. It's not that easy," he said.
"Specific knowledge should always be documented and not in the parameters of the LLM," Andrulis said.
"While fine tuning can be effective at changing a model's style or behavior, it's not the best way to teach it new information."
RAG offers an alternative. The idea here is that the LLM functions a bit like a librarian retrieving information from an external archive.
Collection
[
|
...
]