Over a Third of New Web Text is AI-Generated; Pangram 4 Claims 1-in-24k False Positive Rate
aftahi_ai · x · 2026-08-02
Over a third of all new text on the internet is now generated by machines, including 26% of long-form social posts and 9% of news articles.
To combat this, Pangram lab developed Pangram 4, a detection tool specifically optimized to identify text generated by frontier models like Claude, GPT-5.6, and Gemini. According to the developers, the model was tested across over a million human-written texts and boasts a false positive rate of roughly 1 in 24,000, demonstrating exceptional accuracy.
More from Models
- User Mocks Fable Model: Like a Drunk Crank Physicist, Useless for Math Research — jasondeanlee · 2026-08-02
- Why Multimodal Input Matters for AGI: DeepSeek & Anthropic's Approach — dotey · 2026-08-02
- MiniMax H3 Model Expected to Launch Tomorrow — Xianbao_QIAN · 2026-08-02
- User Reports: Claude Opus Excels at Precise Tasks but Fails as a Daily Driver — EXM7777 · 2026-08-02
- DeepSeek's Tiny Model Size Yields 40x Opus Throughput, Enabling Low Prices — gaganghotra_ · 2026-08-02
- KRAFTON Releases A.X-K2 Speech Model Supporting Any-to-Any Modality — KRAFTON · 2026-08-02