NCM chat (experimental, rough) · Next-Concept Modeling, Stanford Hazy Research (arXiv 2610.14162). Runs fully in your browser: the released XSum NCM generator, briefly fine-tuned on short dialogue, fp16-stored-weights ONNX, fp32 compute, verified identical to the PyTorch model on test prompts. Replies are short and often wrong or garbled. Not a capable assistant. First load downloads the model (~375 MB, needs roughly 1 GB free RAM; desktop browser recommended).