The Oracle no neural network

A compiler-learning model. Every answer carries numbered citations, a rule with its premises, or an honest "I don't know" — and it learns from every sentence it cannot parse. Watch it think on the right.

Try: Who are you? · Summarize Krakatoa. · Compare the Moon and Mars. · Why are the Moon and Mars different? · What if a bird had no wings? · Explain gravity simply. · Give me a timeline of Krakatoa. · Is a beagle an animal? · How does a volcano erupt? · What causes earthquakes? · then: Why did you say that? · Simpler. · Tell me more.

Teach it a document

Paste text; sentences pass the same admission law as the public corpus (4–60 words, clean, terminated — rejected, never repaired). Then ask: "Tell me about <your title>." Session-scoped.




Conversations are recorded to teach the model: every sentence it cannot parse becomes a rule candidate. · the paper · wanderaround.io