Output Diversity

The breadth of distinct outputs an LLM can produce, which is reduced by mode collapse after alignment training

Neighborhood — ranked by edge-count

paper

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Semantic Diversity Scoremethod0.775
Diversity metric computed as 1 minus mean pairwise cosine similarity of response embeddings, using OpenAI's text-embedding-3-small
Diversity Parameterconcept0.772
Gold standard value (e.g., nucleus sampling p-value) used as ground truth for evaluating diversity metrics
Multi-Response Diversityconcept0.770
Diversity measured over a set of m responses for a single conversation, as opposed to test set diversity
Output-truthconcept0.768
The correctness of a model's generated outputs, distinct from the correctness of statements provided as input.
Semantic Diversityconcept0.767
Core concept measured by NLI Diversity — diversity of meaning across a set of dialogue responses
Diversity Thresholdconcept0.759
Target minimum diversity level (e.g., 10 contradictions) that DTG iterates toward
Test Set Diversityconcept0.752
Traditional diversity measurement over one response per conversation across the test set
Diverse Intelligenceframework0.749
Research program studying intelligence at multiple scales and substrates; proposed as relevant to implications of mnemonic improvisation.