I'm Kenyan. I Don't Write Like ChatGPT. ChatGPT Writes Like Me. via Marcus Olang'
This is a great example of preference tuning away from an accurate world model (in this case, tuning by the Kenyan reinforcement trainers on LLMs to their idea of perfect written English).
Reference
-
Permalink (
2026.NTE.041) - On
- In Notes
- Tagged llm, alignment
- From I'm Kenyan. I Don't Write Like ChatGPT. ChatGPT Writes Like Me.
- Edit
| ← Previous | Next → |
| Fast Row 🚣 | Note on Flying With Kids and the Future of Mankind via Hallie Haglund |