# OpenAI — AI-written critiques help humans notice flaws

- Company: OpenAI (openai.com)
- Announced: 2022-06-13T07:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/critiques
- Record: https://forck.live/items/904-ai-written-critiques-help-humans-notice-flaws
- Subject: GPT / ChatGPT / API

OpenAI trained critique-writing models that help humans notice flaws in summaries. Larger models are better at self-critiquing, and scaling improves critique-writing more than summary-writing.

## Evidence

Verbatim from https://openai.com/index/critiques:

> We trained “critique-writing” models to describe flaws in summaries. Human evaluators find flaws in summaries much more often when shown our model’s critiques. Larger models are better at self-critiquing, with scale improving critique-writing more than summary-writing.

---

Record: https://forck.live/items/904-ai-written-critiques-help-humans-notice-flaws
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md