Pangram
Pangram enters the wiki through Taken littorally: Spain’s sudden crisis in Ceuta as an AI-writing detector named by Caitlin Talbot. The source does not review Pangram as a product; it uses it as an example of detector tools that can produce false positives and offer little explanation.
Its wiki role is therefore cautionary. Pangram sits inside AI Writing Detection and AI Detector Bias: detection can help prompt scrutiny, but a score alone is not a reliable authorship judgment without human review, context, and process evidence.
Substack CEO on the platform’s new AI detector adds Pangram as the detector provider behind Substack’s new AI-detection feature. Chris Best says the feature estimates how much text was written by a person or by AI, and that Substack treats false positives as more serious than false negatives because human-written work can be mislabeled as machine-generated.
This source changes Pangram’s wiki role from only a cautionary detector example to a platform-infrastructure case. The detector is still not treated as proof, but it becomes part of a broader disclosure workflow where users can report mistakes and writers can add process statements about AI use.
AI loves negative parallelism adds Pangram as a corpus-measurement source for Negative Parallelism. The episode says Pangram found the “not X, but Y” construction about three times more often in AI-generated prose than in entirely human writing, making Pangram relevant not only as a detector product but also as a source of comparative style evidence.
Connections
- Caitlin Talbot and The Intelligence - source context.
- AI Writing Detection, AI Detector Bias, and AI Writing Pedagogy - concepts connected to detector reliability.
- Human Judgment Under AI - broader responsibility frame.
- Substack, Chris Best, AI Content Provenance, and AI Authorship Presence - platform transparency and reader-expectation branch.
- Will Oremus, The Atlantic, and Negative Parallelism - Marketplace Tech branch on AI-writing tics.