Interactive demo · curated papers · scripted responses · no model calls
Follow your curiosity.
Pick an interest. Explore real papers, follow editorial connections and try a conversation with inspectable notes. No account needed.
Select a node to inspect its paper. Links group editorial themes; graph distance has no scientific score. Use 2D or List for keyboard exploration.
REALM: Retrieval-Augmented Language Model Pre-Training
Guu et al.
REALM brings document retrieval into language-model pretraining. It treats an external collection as accessible knowledge rather than expecting every fact to live in model weights.
Read the curated demo note
Editorial context, not a paper quotation.
REALM learns retrieval through a masked-language objective. Its place in this reading set is the training stage: how a model learns to consult an external collection.
Reading-list and review changes live only on this page. They are not saved to a workspace.
Read the original publicationA small library. A bigger picture.
8 real publications · curated notesTry the conversation
A question opens a reading path.
These responses are prewritten examples. Inspect the attached editorial notes, compare approaches or choose where to read next. No model runs.
Your temporary reading list
0 savedPractice gathering a reading list. It stays in this tab’s memory and disappears on refresh or Reset demo.
Save a paper from its detail panel to start.
What is real in this demo?
Interactive demo: responses are prewritten, no AI model runs, and nothing is saved to a workspace. Papers are real; notes and thematic links are editorial examples, not paper quotations, verified citations or model assessments.
Only topic changes and explicit questions contact the public demo endpoints. Source publications open only when you follow their links. Reading-list and review actions make no network requests.