Visualizers
Learn by moving things
Interactive explainers for the ideas behind retrieval, embeddings, and agents. A formula tells you what is true; dragging a vector shows you why. Each one also appears inside the article it belongs to, and runs entirely in your browser, so drag, poke, and break it.
FLARE: retrieve mid-generation
From: RAG That Thinks →Forward-looking active retrieval woven into generation. As the model writes, confident words stream straight out, but whenever it is about to say something it is unsure of, it pauses, retrieves on that span, and continues.
Confident words are written straight out; retrieval only fires on the shaky ones.
Self-RAG: grade yourself as you write
From: RAG That Thinks →The model emits special reflection tokens that gate its own flow: whether to retrieve, whether a passage is relevant, and whether the answer is actually supported. A 'not supported' verdict sends it back to revise, in-band, before anything is shown.
Special reflection tokens decide whether to retrieve, whether a passage is relevant, and whether the answer is actually supported. A “no” on the last one sends it back to revise, in-band, before anything is shown.
Speculative RAG: draft several, verify, pick
From: RAG That Thinks →A small model drafts several candidate answers in parallel, each from a different retrieved subset; a larger verifier model scores them and picks the best. Parallel drafting plus one verify beats a long serial loop.
A small model drafts several candidate answers in parallel, each from a different retrieved subset of the documents.