Retrieval-Conditioned 4-Gram & Dual-Head Endnote Engine

Prompt a Brain That Only Read Infinite Jest

Ask any question or type any topic below. The engine maps your prompt to matching semantic clusters across the 1,079-page novel using BM25 retrieval, then synthesizes a coherent response using 4-gram clause chaining while autonomously generating matching endnotes.

Try a prompt:

Streaming response...

Coherence & Context ArchitectureActive Engine

Temperature (T) 0.75

Initializing 638,762-token Infinite Jest corpus index...

Autonomous Endnote Head (72,070-Token Trained Sub-Model) 0 Endnotes Spawned

1. Prompt Semantic Attention (BM25) 950 segments indexed

Concepts activated in Infinite Jest by your prompt:

2. Live Next-Token Softmax Probabilities Click Bar to Steer

Why This Is Coherent:

Instead of picking isolated words with a 1-word memory, Stage 4 retrieves the top thematic passages matching your prompt via BM25 and chains 4-gram clauses across shared pivot words.

Empirical Training Telemetry · Extracted From Your Infinite Jest EPUB

Total Book Tokens

638,762

566,692 main + 72,070 endnotes

Unique Vocabulary |V|

37,585

Plus 441 [FN] endnote jumps

Unigram Perplexity

1,263.8

H₁ = 10.30 bits/token

Multi-Gram Perplexity

4.9

H₃ = 2.29 bits/token (258x reduction)