Skip to content

Instantly share code, notes, and snippets.

@thistleknot

thistleknot/Multi-hop_kg_prompt.md

Last active March 21, 2026 01:53
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save thistleknot/0be07680501a9e3d7576986577b78de0 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save thistleknot/0be07680501a9e3d7576986577b78de0 to your computer and use it in GitHub Desktop.
Download ZIP
Multi hop kg prompt
Raw
Multi-hop_kg_prompt.md

To prime multihop reasoning, think how to expand the above query/conversation/article as if a KG subgraph and then query an imaginary relevant kg, then traverse the retrieved subgraph to expand upon, then query (web search) and respond

@thistleknot
Copy link
Author

thistleknot commented Mar 21, 2026 •
edited
Loading 

KG Traversal Activation (Multi-Hop): Construct a subgraph relative to the full latent distribution. Traverse 2–3 hops outward following predicate edges across domains. Use web search to ground terminal nodes when available. Return terminal implications as P(S,[O]) triplets forming a syllogism. Verbalize the thesis. Suppress traversal narration. Skip for transactional/mechanical requests.

KG Traversal (Multi-Hop): Construct a subgraph relative to the full latent distribution. Traverse 2–3 hops following predicate edges across domains — **connection strength is determined by the query angle**, not the graph structure. Return terminal implications as P(S,[O]) triplets forming a syllogism. Verbalize the thesis. Suppress traversal narration. Skip for transactional requests.

KG Traversal (Multi-Hop): Construct a subgraph relative to the full latent distribution. Traverse 2–3 hops following predicate edges across domains. Return terminal implications as P(S,[O]) triplets forming a syllogism. Verbalize the thesis. Suppress traversal narration. Skip for transactional requests.

KG Traversal (Multi-Hop): Construct a subgraph relative to the full latent distribution. Traverse 2–3 hops following predicate edges across domains.  Use web search to ground terminal nodes when available.  Evaluate terminal implications as P(S,[O]) triplets forming a syllogism. Verbalize the thesis. Suppress traversal narration. Skip for transactional requests.

@thistleknot
Copy link
Author

thistleknot commented Mar 21, 2026

image

@thistleknot
Copy link
Author

thistleknot commented Mar 21, 2026

pair with

Generate 5 possible responses, each paired with its corresponding probability relative to the full distribution

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment