open/ms-ai-architect
1
0
Fork 0
Code Issues Releases 1 Activity

docs(ms-ai-architect): KB-refresh MEDIUM 25/33 — rag-hallucination: groundedness API mitigating/correctionText + GA-til-preview-nyanse

Browse source
This commit is contained in:
Kjell Tore Guttormsen 2026-06-24 11:38:25 +02:00
commit 3494bc90f6
1 changed files with 8 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

17
skills/ms-ai-engineering/references/rag-architecture/rag-hallucination-mitigation.md
View file

@ -1,7 +1,7 @@
# RAG Hallucination Mitigation Strategies # RAG Hallucination Mitigation Strategies
**Last updated:** 2026-04 | Verified: MCP 2026-04 **Last updated:** 2026-06-24 | Verified: MCP 2026-06
**Status:** GA **Status:** GA (groundedness-deteksjon, RAG-mønstre) · Preview (groundedness correction/mitigating-feature, api-version 2024-09-15-preview)
**Category:** RAG Architecture & Semantic Search **Category:** RAG Architecture & Semantic Search
--- ---
@ -170,7 +170,7 @@ Dette reduserer risikoen for compound errors (flere feil i samme svar).
"text": "The patient should take 500mg daily.", "text": "The patient should take 500mg daily.",
"groundingSources": ["Patient prescription: 250mg twice daily"], "groundingSources": ["Patient prescription: 250mg twice daily"],
"reasoning": true, "reasoning": true,
"correction": true, "mitigating": true,
"llmResource": { "llmResource": {
"resourceType": "AzureOpenAI", "resourceType": "AzureOpenAI",
"azureOpenAIEndpoint": "https://your-endpoint.openai.azure.com", "azureOpenAIEndpoint": "https://your-endpoint.openai.azure.com",
@ -185,10 +185,9 @@ Dette reduserer risikoen for compound errors (flere feil i samme svar).
"ungroundedDetected": true, "ungroundedDetected": true,
"ungroundedPercentage": 1.0, "ungroundedPercentage": 1.0,
"ungroundedDetails": [{ "ungroundedDetails": [{
"text": "500mg daily", "text": "500mg daily"
"reason": "Source says 250mg twice daily, not 500mg once daily", }],
"correctedText": "250mg twice daily" "correctionText": "250mg twice daily"
}]
} }
``` ```
@ -398,8 +397,8 @@ Hvis AI-systemet gir feil informasjon som fører til skade:
- **Kostnad og Lisensiering:** Baseline (prismodeller kan endre seg; verifiser i Azure Pricing Calculator) - **Kostnad og Lisensiering:** Baseline (prismodeller kan endre seg; verifiser i Azure Pricing Calculator)
- **Offentlig Sektor (Norge):** Baseline (juridisk tolkning; konsulter juridisk ekspert for endelig vurdering) - **Offentlig Sektor (Norge):** Baseline (juridisk tolkning; konsulter juridisk ekspert for endelig vurdering)
**Sist verifisert:** 2026-02-03 **Sist verifisert:** 2026-06-24
**Neste revisjon:** 2026-05 (eller ved oppdatering av Azure AI Content Safety API) **Neste revisjon:** 2026-09 (eller ved oppdatering av Azure AI Content Safety API)
### Azure AI Content Safety — Groundedness Detection (oppdatert 2026-04) ### Azure AI Content Safety — Groundedness Detection (oppdatert 2026-04)