mixture of experts
for model scaling engineers
Why this scored 10
every term, weightedIts fastest-moving item, measured against the pace of its own source
Whether that velocity is itself speeding up, as a per-hour rate
How many independent communities its own items come from
Decays to zero over 14 days, counted from when we first saw it
Subtracted once something is big and old — sized by its biggest item, aged from when we first saw it
Weights are hand-tuned, not learned — we're calibrating them against realized trends as history accumulates. On an entity's first sighting there's no previous reading to compare against, so acceleration starts from a neutral prior rather than a measurement, and velocity falls back to engagement over its whole lifetime until a second reading exists. Full methodology
Signal history
7-day window (free)The evidence
The live items this entity's score aggregates — every community independently talking about it right now. This is the corroboration, shown, not claimed.
- 119
Sparsity Saves Compute, Not Samples: Oracle Inequalities for Dense and Sparse Mixture of Experts
Shows sparsity reduces compute cost while dense experts need more samples. · for model scaling engineers
arxivSteadyai48m ago - 218
ID Balancing: Stable Training of Extremely Sparse MoE via PID-Based Load Control
PID load control stabilizes training of extremely sparse MoE models. · for ml engineers, ai researchers
arxivSteadyai13d ago