Users express optimism about SwiGLU proposed as fast weights alongside NVP and MoE layers because it suggests progress toward better abstractions for scaling intelligence.
Based on 2 visible X reactions from 3 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@francoisfleuret Fast weights, MoE, SwiGLU… feels like we're still discovering the right abstractions for scaling intelligence.
4:39 AM · Jul 15, 2026Users express optimism about SwiGLU proposed as fast weights alongside NVP and MoE layers because it suggests progress toward better abstractions for scaling intelligence.
Based on 2 visible X reactions from 3 accounts; directional sample.
Ask a question below.
Published answers will appear here.