Skip to content

1) Summarization (enterprise)

Goal

Produce concise, faithful summaries with low variance.

Typical risks

  • hallucinated facts
  • missing critical details
  • overlong output
  • temperature: 0.0–0.3
  • top_p: 1.0 (tune only if you choose nucleus sampling)
  • presence_penalty: 0
  • frequency_penalty: 0–0.3 (optional)
  • max_tokens: set pretty low (e.g., 128–512) depending on policy

Strong pattern: structured summary schema

Use response_format with json_schema so your summary always includes the fields leadership cares about.

Example schema:

{
  "name": "Summary",
  "strict": true,
  "schema": {
    "type": "object",
    "additionalProperties": false,
    "properties": {
      "title": {"type": "string"},
      "key_points": {"type": "array", "items": {"type": "string"}},
      "risks": {"type": "array", "items": {"type": "string"}},
      "next_steps": {"type": "array", "items": {"type": "string"}}
    },
    "required": ["title", "key_points", "risks", "next_steps"]
  }
}

How to test

Sweep temperature and max_tokens first; measure:

  • summary length compliance
  • JSON validity (if structured)
  • human faithfulness score

Related docs: Parameters → Decoding, Structured Outputs; Experiments → Presets.