New Felda writing model claims to pass every AI detector with quality prose
almmaasoglu · x · 2026-09-04
almmaasoglu claims Felda's latest writing model can produce, edit, and continue complete pieces of 500-4,000 words that every AI detector they tested classifies as human — while delivering strong prose. The author says detection evasion wasn't the goal: it emerged naturally from optimizing for structural coherence across long passages (where even frontier models often fail) and natural shifts in language and rhythm. Unverified self-report; third-party testing pending.
More from Models
- GLM-5.3 Flash gives 6x more tokens per dollar than Gemini 3.8 Flash in real test — MaziyarPanahi · 2026-09-04
- LangSmith data: gpt-4o-mini reaches 13% of orgs; DeepSeek V4 Flash is the only open-weight model on the lists — LangChain · 2026-09-04
- Polymarket puts 60% odds on next Claude Opus releasing by end of September — Polymarket · 2026-09-04
- Gemini 3.8 Flash Lands on Netlify AI Gateway Day One: Zero Config, No Keys, Built-in Caching — thisiskp_ · 2026-09-04
- DLSS 5 looks 'crazy good' as NVIDIA's AI upscaling impresses users — ssh4net · 2026-09-04
- Dev's accidental A/B test: reverting from .8 to .6 reveals dramatic capability gap — AI_Andrew · 2026-09-04