3 ms·

Mostik.ai – latent communication between AI models

• po_westnet26 20d ago
  [flagged]
• arpperzhao 19d ago
  Conjecture: Does GLM only handle the prefill stage, compress its output hidden vectors (trained?), and then send them to Qwen for decoding?