16 November 2026
For the first time, bots sent more traffic to web pages than people did. Cloudflare's CEO put the split at 57.5% bots to 42.5% humans in June 2026, a year earlier than he'd predicted. And until now, every machine-written page had a speed limit: AI writes one word at a time. A Stanford paper just tested a way around that limit. Next-Concept Modeling (NCM) sketches a whole passage first, then fills in the detail level by level. In the paper's tests it ran 10–31× faster than diffusion models of the same size, and solved 86.1% of hard Sudoku boards. A word-by-word model of the same size solved 1%. So what happens to an internet that's already mostly bots when writing gets that much faster and cheaper? And what's still unproven? NCM has only been tested on small research models, and independent tests show the old way still wins when a data center batches many requests at once. CHAPTERS 0:00 Bots just passed humans online 1:17 The flood: what the bot reports found 3:14 AI's hidden speed limit 4:26 More text than ever, and the diffusion detour 5:23 Stanford's Next-Concept Modeling 7:40 The results: speed, Sudoku and sixteen drafts 9:11 Why speed is the real multiplier 10:45 Model collapse and the black box 12:40 Reality check: what NCM hasn't proven 14:10 The human frontier
🚀 Introducing Next-Concept Modeling (NCM): our new paradigm for fast, parallel, coarse-to-fine text generation. Compared to strong diffusion language models (DLMs) at similar parameter size, NCM achieves: ⚡ Up to 31× faster generation on downstream tasks like summarization and machine translation with comparable or better quality. 🏎️ 10–21× faster unconditional generation across TinyStories, LM1B, and OpenWebText, with comparable or better generative perplexity and using 3–5× fewer training tokens. 🧩 Better performance on reasoning tasks like Sudoku, outperforming both AR and diffusion baselines with fewer sampling steps. With @ishanskhare, @IddahMlauzi, @ekellbuch, @simran_s_arora, @Azaliamirh, and @HazyResearch. More details in the thread 🧵👇
— Hermann Kumbong (@KumbongHermann) October 14, 2026