diff --git a/blogs/ai_space_escape.md b/blogs/ai_space_escape.md index 19e1fb7..c3f9c54 100644 --- a/blogs/ai_space_escape.md +++ b/blogs/ai_space_escape.md @@ -72,7 +72,7 @@ Static evaluations, such as MMLU, Spider, and HumanEval, offer capability-specif To address the challenges, Game Arena: -- An incentivized, dynamic benchmark using live computer games to evaluate the interactive and strategic aspect of many other reasoning tasks required in real-life. +- Introduces an incentivized, dynamic benchmark using live computer games to evaluate the interactive and strategic aspect of many other reasoning tasks required in real-life. - Involves three reasoning games, each target at different reasoning capabilities. @@ -88,7 +88,6 @@ To address the challenges, Game Arena: Stay tuned and we will release more details about how we generated the leaderboard! 🔔 - --- ## Results @@ -99,7 +98,7 @@ Stay tuned and we will release more details about how we generated the leaderboa Our evaluation results show the following findings: -- Models with strong reasoning capabilities and multi-turn instruction-following capabilities, such as **claude-3.5-sonnet**, **gemini-2.0-flash-thinking-exp-01-21**, and **gemini-1.5-pro** are ranked high in GameArena. +- Models with both strong reasoning capability and multi-turn instruction-following capability, such as **claude-3.5-sonnet**, **gemini-2.0-flash-thinking-exp-01-21**, and **gemini-1.5-pro** are ranked high in GameArena. - Models that excel at short conversations but with poor reasoning in extended game sessions, such as **Mistral-Large-2**, usually rank low in GameArena (not shown). @@ -118,7 +117,7 @@ Our evaluation results show the following findings: | grok-2-beta | 994.91 | 977.67 | 973.24 | 981.94 | - as of Feburary 2025. For more fine-grained analysis with detailed metrics, please check our paper.") + as of Feburary the 10th, 2025. For more fine-grained analysis and other detailed evaluation metrics, please check our paper.") ### User Tests @@ -157,7 +156,7 @@ We also invite you to try out [game engine](https://github.com/lmgame-org) and b ## Acknowledgement -We would like to thank Yonghao Zhuang, Xin (Kris) Gao, Yinmin Zhong, Yao Fu, Anastasios Angelopoulos, Roger Wang for providing insightful feedbacks. +We would like to thank Yonghao Zhuang, Xin (Kris) Gao, Yinmin Zhong, Yao Fu, Anastasios Angelopoulos, Roger Wang, Ligeng Zhu for providing insightful feedbacks. --- diff --git a/blogs/backup_ai_space_escape.md b/blogs/legacy/backup_ai_space_escape.md similarity index 100% rename from blogs/backup_ai_space_escape.md rename to blogs/legacy/backup_ai_space_escape.md diff --git a/favicon.ico b/favicon.ico index c31b32b..2290f5c 100644 Binary files a/favicon.ico and b/favicon.ico differ diff --git a/index.html b/index.html index 945eeeb..c4d1ace 100644 --- a/index.html +++ b/index.html @@ -1 +1 @@ -