Update card

README.md

@@ -17,5 +17,4 @@ Small models, narrow tasks. Trained from scratch, under 100M parameters, each do
 interpretability lab: watch the models think, one forward pass at a time.
 
 Every number behind them is reproducible: [bg-eval](https://github.com/iksnerd/bg-eval) re-runs
-the published results from the checkpoints alone, with no dependency on this project's training
-code.
+the published results without trusting us.