You don’t have to customize Obsidian, I haven’t customized mine at all. No settings and no plugins. I have the following folder structure and it has been working for me for years.

I trained an image model to generate brand-consistent illustrations. Here is "man typing on laptop" in the style of a Slack illustration, based on just five training examples and 7 minutes of training. Should I build a web interface for training models like this?

I just bought two domains. But which one is better? #buildinpublic

My app focuses on teaching high-impact, common words first. Words are taught in context with example sentences written by native speakers, then reviewed at memory-optimal intervals. Practice any way you like - translating, listening, speaking, flashcards - it's up to you.

Nothing more relaxing than adding a new feature

"Download Llama 70B" is the new "Download Wikipedia" for bootstrapping civilization if the world ends

Infodiet hygiene - potential product idea? MVP: a Chrome extension that picks out all <article> tags and blurs or hides them if they fail a classifier test.

RAG (summarized): - Chunk your text - Use a vector store instead of a brute force algorithm - Use a 2-level hierarchy, first searching summaries and then searching documents - Use a LLM to generate hypothetical questions for each chunk, and search for that. Flipside: ask an LLM to generate a hypothetical response for a user query and search with that vector too. - Split your documents into sentences for finegrained searches, and add surrounding sentences as context when sending the match to the LLM. - Alternatively, split your document into chunks; extract the sentences, and send the whole chunk to the LLM if you have >k matches for sentences in the chunk to a query. - Use an old school keyword search algorithm like BM25 to enrich your results - Use a LLM to transform the query into something more general, when there aren’t enough valuable matches for a specific query - Use fuzzy matching to figure out which parts of a response reference which sources - Use a LLM to specify which particular data store(s) to search in - Use a LLM to summarise and select which search results are relevant