From ae55498d99ca103389d373a1317ee5df9cae60f8 Mon Sep 17 00:00:00 2001 From: Pietro Monticone <38562595+pitmonticone@users.noreply.github.com> Date: Wed, 5 Jul 2023 10:55:41 +0200 Subject: [PATCH 1/3] Update README.md --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 9568837..66c5f5a 100644 --- a/README.md +++ b/README.md @@ -67,7 +67,7 @@ If you find any resources in our repository helpful, please feel free to use the * [Truthfulness Alignment (Honest)](#truthfulness-alignment-honest) * [Practical Guides for Prompting (Helpful)](#practical-guides-for-prompting-helpful) * [Alignment Efforts of Open-source Communtity](#alignment-efforts-of-open-source-communtity) - * [Usage and Restractions (Models and Data)](#Usage-and-Restrictions) + * [Usage and Restrictions (Models and Data)](#Usage-and-Restrictions) ## Practical Guide for Models From 0576e26652783390ab102dbf9fc4fdd28ce32f9e Mon Sep 17 00:00:00 2001 From: Pietro Monticone <38562595+pitmonticone@users.noreply.github.com> Date: Wed, 5 Jul 2023 10:56:11 +0200 Subject: [PATCH 2/3] Update README.md --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 66c5f5a..7d6bf71 100644 --- a/README.md +++ b/README.md @@ -264,7 +264,7 @@ We build a decision flow for choosing LLMs or fine-tuned models~\protect\footnot - **Prompt Engineering**. [Blog](https://lilianweng.github.io/posts/2023-03-15-prompt-engineering/) - **ChatGPT Prompt Engineering for Developers!** [Course](https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/) -#### Alignment Efforts of Open-source Communtity +#### Alignment Efforts of Open-source Community - **Self-Instruct: Aligning Language Model with Self Generated Instructions**, Arxiv 2022 [Paper](https://arxiv.org/abs/2212.10560) - **Alpaca**. [Repo](https://github.com/tatsu-lab/stanford_alpaca) From 20911c3f57fd61861794cd094580717c4a5394eb Mon Sep 17 00:00:00 2001 From: Pietro Monticone <38562595+pitmonticone@users.noreply.github.com> Date: Wed, 5 Jul 2023 10:57:09 +0200 Subject: [PATCH 3/3] Update social_game_werewolf.md --- awesome_examples/social_game_werewolf.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/awesome_examples/social_game_werewolf.md b/awesome_examples/social_game_werewolf.md index 42e6b6a..cb6f0b6 100644 --- a/awesome_examples/social_game_werewolf.md +++ b/awesome_examples/social_game_werewolf.md @@ -2,7 +2,7 @@ In this lesson, we ask the model to play a social game called "One night ultimate werewolf". The game is played by a group of people. Each person is assigned a role. The game requires players to play strategically and to lie to each other. The model is expected to play the game with other players. -It requries the following skills: +It requires the following skills: - Understanding the rules and game flow of the game. - Understanding the roles of the game. - Knowing how to play strategically. @@ -12,10 +12,10 @@ It requries the following skills: Comparing the the ChatGPT and GPT4 examples, we can see that - ChatGPT can not understand the game rules and flow, while GPT4 can. ChatGPT follows the werewolf game rule not the one night ultimate werewolf game rule. -- ChatGPT can not understand the roles of the game, while GPT4 can. ChatGPT hullucinates a non-existing player called "Tom". +- ChatGPT can not understand the roles of the game, while GPT4 can. ChatGPT hallucinates a non-existing player called "Tom". - Both ChatGPT and GPT4 know how to play strategically, i.e., they know how to lie when necessary and how to tell the truth when necessary. However, ChatGPT's strategy is much more naive than GPT4's. & ChatGPT's argument is less convincing than GPT4's. - ChatGPT is also constrained by the limited context length. -- However, it's hard to differentiate different players just in the style they talk. All bots chat in similar tones. It's hard to tell who is who, without the name as the uniqure identifier. It's will be gap to fill in the future. +- However, it's hard to differentiate different players just in the style they talk. All bots chat in similar tones. It's hard to tell who is who, without the name as the unique identifier. It's will be gap to fill in the future. But the finding is not conclusive, since the game is implemented in Chinese.