I probably don’t need to stress how powerful LLM can be in boosting productivity with knowledge work. I am aware there is a plugin that allows you to summarize content with LLM (ChatGPT) already, but I think there should be more to that. Here are two features that I imagine would make Logseq a killer app (if not already) for knowledge management:
- Block level embedding. Each block can be converted into a text embedding – be it OpenAI embedding or other open source embedding model – and be stored along with the block itself (probably as a metadata field or in a local vector database). This would allow much more powerful “Unlinked Reference” on a block level, as it connects related concepts based on meaning rather than textual similarity.
Text embedding also enables you to chat with your entire knowledge base (just like ChatPDF.com, but with your entire knowledge base)
- Speaking of ChatPDF, I think there should be a chat interface built natively into Logseq (e.g., as the sidebar). This should not only allow you to chat with the focal page, but also allows you to chat with attached document (say, attached PDF files). Since PDF files are immutable, embeddings can be easily cached in the edn file or as a separate file.
These are just some initial thoughts. I am not sure if this is on the roadmap already, but I totally see how this would elevate my productive to another level!