Building a 'deslop writing' skill proves near impossible; fine-tuning may be the only reliable fix
AccBalanced · x · 2026-11-15
BenjDicken spent an afternoon trying to build a "deslop writing" skill and found it far harder than expected. Approaches tried: explicitly instructing the model to edit common AI-slop patterns; hill-climbing with fable 5.2 using Opus as writer and Pangram as oracle; building a voice profile from his past writing; testing existing skills that claim to do this. A screenshot shows a 2,230-word article still flagged 97% AI by Pangram. His conclusion: without a human in the loop, the only reliable route is fine-tuning a model on a specific human's tone and work — quipping that human-audience writing may be the final-frontier job.
More from coding & agent
- Can AI invent a new competitive game genre in 4 hours? 18 prototypes so far — gabriel1 · 2026-11-15
- iceberg-mcp-server: an open MCP server to query and manipulate Apache Iceberg tables — dragonejt · 2026-11-15
- Kyvenza 5.0 turns Mac VMs into AI agent workspaces with MCP file, shell and mouse tools — ScholarBackground836 · 2026-11-15
- A robot band in your browser: physics-simulated instruments you can play via MIDI — Active_Cauliflower71 · 2026-11-15
- Tether's QVAC BotPsy-0.8B Doubles Tool-Use Scores, Beats Models 2.5x Its Size — NielsRogge · 2026-11-15
- Claude writes 12,500 lines of Blender Python to make a 4K film, renders with 89 rented GPUs for $20 — john_lam · 2026-11-15