Show HN: How clanker are you? A reverse Turing test
13 hours ago
- #Surprisal measurement
- #Writing analysis
- #AI detection
- A tool called "How clanker" analyzes Twitter (X) posts to measure how AI-like (predictable) someone's writing is.
- It pulls recent posts from a user, excluding retweets and replies, and scores each word based on its surprisal (predictability) under language models.
- Low surprisal (high predictability) results in a "clanker" (AI-like) score, while high surprisal indicates more human-like writing.
- The method uses log probabilities from models, normalizing scores against baselines, and the overall score reflects the nearest model's assessment.
- In demo mode, scores are simulated using a deterministic stand-in instead of real models, and the tool is presented as fun, not a judgment of the person.