LLMs can easily produce data not in training dataset.

LLMs do not navigate stored data. An LLM is not a DB of the training data.

I've had the same thought as above but unfounded (just a feeling, pretty much) so I'm curious to learn more. Do you have any references I can check out that supports these claims?

Come up with a novel puzzle that is guaranteed to not be in the training set, and ask GPT-4 to solve it.

To control for that doesn't seems trivial.

But indexing *is* training. It's just not using end-to-end gradient descent.

You might like, the Infinigram paper then. It was discussed recently:

https://news.ycombinator.com/item?id=40266791

RAG (Retrieval-Augmented Generation) enhances Large Language Models (LLMs) by integrating them with specific domain knowledge or an enterprise's internal databases, eliminating the need for model retraining. This method maintains the relevance, accuracy, and value of the outputs in a cost-effective way. However, the effectiveness of a RAG solution heavily depends on the quality of the datasets it uses. These datasets often come from varied sources and in different formats. For optimal performance, it's crucial to normalize the various data types, clean and organize the data by removing unnecessary elements such as special characters, irrelevant metadata, or extraneous text.

HelloRAG.ai addresses this challenge by offering a no-code platform that allows you to easily transform multi-format datasets into structured formats like JSON. This tool helps you focus on developing the logic of RAG, streamlining the process of leveraging augmented data for enhanced model performance

The models are multiple orders of magnitude smaller than the compressed versions of their training data, they can not be the equivalent of a DB of it.

The training data is ideo-semantically compressed? News to me... is it perhaps stored in kanji?