#econsky
#techsky
#Meta dropped #MuseSpark: ~72B params, not SOTA, but maybe ~10x compute-efficient on reasoning.
1. Why closed-source "small"? Eff., speed, IP. Less Llama-style scale chasing, more deployable #AI.
2. Biz opp: #paidAPI + on-device AI for Meta's users.
Right 1st step to #SOTA?