Custom harness with abliterated DeepSeek v4.1 fools Pangram 4.1, scoring 80–100% "human written"
IndraVahan · x · 2026-11-19
IndraVahan reports beating the Pangram 4.1 AI-text detector again with a custom writing harness, this time swapping in an abliterated DeepSeek v4.1 flash 1017. After a few optimizations, single-shot outputs scored 80–100% "human written" labels across every sample, with a screenshot showing a 100% Human Written verdict. His takeaway: evading current AI-text detection is far too easy.
More from Fun
- Michele Parrinello's twin brother photo sparks 'symmetry breaking' meme — mmbronstein · 2026-11-19
- AI race meme updated to "November 18, 2026" — fictional model versions keep battling — m3nt3_ · 2026-11-19
- "It's one prompt away": skeptic says AI launch dates are engineered for vibes, not magic — ryunuck · 2026-11-19
- Meta-analysis of 912 neuroimaging studies finds no fixed neural substrate for the self — dioscuri · 2026-11-19
- Joke: LinkedIn will lose it when GTA VI launches next week — tech__unicorn · 2026-11-19
- 'Artificial' review: Guadagnino's OpenAI drama is a sprawling mess — ShakeelHashim · 2026-11-19