Update README.md

README.md

@@ -26,7 +26,7 @@ configs:
 
 # PoolBench
 
-**PoolBench** is a benchmark for evaluating 19 hidden-state pooling strategies for activation steering across 17 text concepts and 7 language models.
+**PoolBench** is a benchmark for evaluating pooling strategies for activation steering across 17 text concepts.
 
 Each concept corpus contains up to 1,000 passages per class (700 train / 300 test) sourced from real public datasets on HuggingFace. No passages are LLM-generated.
 
@@ -57,6 +57,5 @@ Each JSON line has the following fields:
   title={PoolBench: Evaluating Pooling Strategies for Activation Steering Vectors},
   author={Anonymous},
   year={2026},
-  note={NeurIPS 2026 Datasets and Benchmarks Track}
 }
 ```