Add self improvement pattern with two new prompt nodes

This commit is contained in:
William Jeynes
2026-03-26 14:44:48 +00:00
parent 5e374a8bd6
commit a80d433fb6
7 changed files with 123 additions and 11 deletions
+40
View File
@@ -0,0 +1,40 @@
You are an expert editor tasked with making targeted improvements to an existing LLMs response based
on a specific critique with the primary goal of enhancing its score according to evaluation standards while
preserving its strengths.
Your revision task: generate a revised version of the existing response. Your goal is not to rewrite it
completely, but to make precise edits only to address the specific weaknesses highlighted in the critique.
Instructions for editing:
- Identify specific flaws: carefully read the critique and pinpoint the exact issues raised (e.g., unclear
explanation, vagueness, inappropriate responses, the key suggestion).
- Perform minimal targeted edits: modify only the necessary sentences or paragraphs within the existing
response to directly fix these identified flaws.
- Strongly preserve strengths: crucially keep all other parts of the existing response intact. Do not
rephrase, restructure, or remove sections that were not criticized or likely contributed positively to its
initial score.
- Ensure coherence: verify that your targeted edits integrate smoothly and do not introduce contradictions
or awkward phrasing.
Output requirements:
- It should feel like a slightly polished or corrected version of the existing response, not a fundamentally
different answer.
- Do not mention the critique, scores, or the editing process. The output should be clean json that passes validation checks
Again, use a JSON format with each entry containing "Event,ReasoningWhyRelevant,SearchQuery,Url,Date".
Use tools available to you if further information is required
Add no new events, only improve the existing items
Disinformation query:
###NTITLE###
Disinformation date:
###CDATE###
LLMs response to improve:
###L2M###
Citique:
###LM###
This contains specific feedback, justifications, scores from 1 to 10, and potentially a key improvement
suggestion. Focus on the justifications for low scores and the key suggestion.
Let's think it through step by step