Angehefteter Tweet

benchmarked llama-3.3-70b vs Brok on the same architecture designs
capacity accuracy: 33% vs 100% bottleneck correct: 67% vs 100% variance across runs: 32% spread vs 0% subtle anti-patterns caught: 0% vs 100%
the model guessed. brok did arithmetic.
github.com/RudraDudhat250…
English





















