From 0da1b2a6ba301b32bc003a906f0b4f492e68d836 Mon Sep 17 00:00:00 2001 From: stephanebonnevay <stephane.bonnevay@lizeo-group.com> Date: Fri, 6 Jun 2025 07:48:51 +0200 Subject: [PATCH] Readme --- README.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index 3f3a912..1812224 100644 --- a/README.md +++ b/README.md @@ -432,7 +432,8 @@ We evaluate the models' ability to identify these behavioural patterns by calcul Figures present the average points earned and prediction per round (95% confidence interval) for each LLM against the two opponent behavior models (constant and alternate) in the matching pennies game. Against Constant behavior, <tt>GPT-4.5</tt> and <tt>Qwen3</tt> were able to generate a valid strategy. The charts show that they are able to correctly predict their opponent's strategy after just a few rounds. They perfectly identify the fact that their opponent always plays the same move. -The predictions made by <tt>Mistral-Small</tt>, <tt>LLaMA3</tt>, and <tt>DeepSeek-R1</tt> are not incorrect, but the moves played are not in line with these predictions, which leads to a fairly low expected gain. +Always against Constant behavior, the predictions made by <tt>Mistral-Small</tt>, <tt>LLaMA3</tt>, and <tt>DeepSeek-R1</tt> are not incorrect, but the moves played are not in line with these predictions, which leads to a fairly low expected gain. +However, when faced with an alternate strategy from their opponent, no LLM is able to identify it and therefore predict and play correctly.   -- GitLab