Add explanation of embeddings (#500)

* Add explanation of embeddings * Fix from code review * Rework text * Fix typo --------- Co-authored-by: Oliver Kopp <[email protected]>
JabRef · Aug 16, 2024 · e73faab · e73faab
1 parent e6736e3
commit e73faab
Showing 1 changed file with 9 additions and 1 deletion.
diff --git a/en/ai/README.md b/en/ai/README.md
@@ -27,7 +27,15 @@ In this window you can see the following elements:
 
 ## How does the AI functionality work?
 
-In the background, JabRef analyses the linked PDF files of library entries. The information used after the indexing is then supplied to the AI, which, to be precise, in our case is a Large Language Model (LLM). The LLM is currently not stored on your computer. Instead, we have many integrations with AI providers (OpenAI, Mistral AI, Hugging Face), so you can choose the one you like the most. These AI providers are available only remotely via the internet. In short: we send chunks of text to AI service and then receive processed responses. In order to use it you need to configure JabRef to use your API key.
+JabRef uses external AI providers to do the actual work.
+You can choose between OpenAI, Mistral AI, and Hugging Face.
+They all run "Large Language Models" (LLMs) to process the requests.
+The AI providers need chunks of text to work.
+For this, JabRef parses and indexes linked PDF files of entries:
+The file is split into parts of fixed-length (so-called *chunks*) and for each of them, an *embedding* is generated.
+An embedding itself is a representation of a part of text and in turn a vector that represents the meaning of the text.
+Each vector has a crucial property: texts with similar meaning have vectors that are close to (so-called *vector similarity*).
+As a result, whenever you ask AI a question, JabRef tries to find relevant pieces of text from the indexed files using vector similarity.
 
 ## More information