@gerardsans
Your LLM is a 40-ton truck screaming downhill. Context is your cargo. Mid ride, good luck steering token drift. RL, CoT, ICL, scaling… can’t fight attention bottlenecks or leave pre-training main roads. Yet another reason for a post-transformer design: https://t.co/lh1TcX0kFY https://t.co/i27Vi1YpSU