From 4bbd1e919e3f11a28749290a4a458db4f606ee1a Mon Sep 17 00:00:00 2001
From: Siavash Golkar <35383824+golkar@users.noreply.github.com>
Date: Sun, 16 Jun 2024 14:17:35 -0400
Subject: [PATCH] Update 2024-05-30-counting.md

---
 _posts/2024-05-30-counting.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/_posts/2024-05-30-counting.md b/_posts/2024-05-30-counting.md
index b45aec8..68f76fd 100644
--- a/_posts/2024-05-30-counting.md
+++ b/_posts/2024-05-30-counting.md
@@ -50,8 +50,8 @@ This information helps disambiguate the different regions based on context.
 
 #### Key Propositions
 
-1. **Proposition 1:** If the regional contextual position information is available in the latent representation of the tokens at some layer of a Transformer, the contextual counting task can be solved with a single additional layer.
-2. **Proposition 2:** A causal Transformer with a single layer and no position encoding (NoPE) can infer the regional contextual position.
+- **Proposition 1:** If the regional contextual position information is available in the latent representation of the tokens at some layer of a Transformer, the contextual counting task can be solved with a single additional layer.
+- **Proposition 2:** A causal Transformer with a single layer and no position encoding (NoPE) can infer the regional contextual position.
 
 These propositions imply that a two-layer causal Transformer with NoPE can solve the contextual counting task.