fix(moderation prompt): lower false positives in toxic category

[mantagen]

Description

I tried a few variants locally (prompt/model combinations), and this was the best I found. (note: gpt-4o-mini is no good for this!).

• updates open-ai from 4.52.0 to 4.58.1 (allowing for json_schema mode)
• moderation agent now uses "json_schema" mode
• adds a 'note' to the Toxic moderation category description
• amends 'instruction' setting of the prompt
• makes justification non-optional
• filter categories by parent category score

Later will explore some more in-depth changes, but that would come with more risk associated, so want to do that when there's a dataset (and more time) to give the confidence required.

Issue(s)

https://www.notion.so/oaknationalacademy/Fix-moderation-bug-where-all-categories-get-flagged-3c93a1e819bb438f971187cb8be653bc

How to test

1. Moderation should generally work (e.g. test some sensitive and toxic material)

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
oak-ai-lesson-assistant	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Sep 10, 2024 9:58am

[github-actions]

Playwright e2e tests

Job summary

Download report

To view traces locally, unzip the report and run:

npx playwright show-report ~/Downloads/playwright-report

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarCloud

[codeincontext]

Makes sense to me 👍

[oak-machine-user]

🎉 This PR is included in version 1.7.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀