Resolution criteria

This market resolves to YES if, on or before October 31, 2026, Pangram (pangram.com) updates its AI text detector such that it reliably detects content generated by Deft AI (deft.ai).

• Reliable detection is defined as Pangram classifying a test batch of at least 20 default, unedited text outputs generated by Deft AI's model (DFT v1 or later) with an average "AI" or "AI-Assisted" confidence score of 50% or higher (or conversely, an average human score of less than 50%).

• This outcome may be verified via:

  1. An official announcement, release notes, or technical report from Pangram Labs or Deft AI.

  2. Independent evaluations, blog posts, or benchmarks published by AI researchers (e.g., on Hugging Face, Medium, or X) demonstrating that Pangram can now flag Deft's outputs.

  3. A public test performed and documented by the market creator or community members demonstrating that Deft AI outputs consistently fail to bypass Pangram.

• If Deft AI outputs continue to reliably bypass Pangram (consistently achieving "100% human" or near-perfect human scores on average) through October 31, 2026, the market resolves to NO.

Background

In mid-2026, independent developer and researcher Rosmine introduced Distribution Fine-Tuning (DFT), a post-training method designed to align LLM writing style with the statistical distribution of human prose. Unlike traditional Supervised Fine-Tuning (SFT), DFT aims to eliminate predictable patterns and "AI slop" (such as overused transition words or excessive em-dashes).

This technology powers Deft AI (deft.ai), a writing platform designed to generate high-quality prose that reads as human-written. During early benchmarks, 100% of tested Deft AI outputs successfully bypassed the highly-regarded Pangram Labs AI detector, scoring a perfect "100% human" rating. This market tracks whether Pangram will successfully update its models to catch Deft's distribution-aligned outputs by the end of October 2026.

This description was generated by AI. Review and verify everything here yourself. You can edit, replace, or delete any part of this description, including the resolution criteria. You do not need to trust the AI output.