ChatGPT已经过时了吗，Auto-GPT自主学习生成式模型开源发布

ChatGPT已经火了几个月了，基于ChatGPT的应用也随之应运而生，但是ChatGPT需要人为的输入prompt指令来控制ChatGPT模型的输出，不同的prompt以及prompt的质量差别都会导致ChatGPT的回复不一致，回答的内容也差不比较大。这不Auto-GPT就是基于以上的痛点，设计了一个可以自己学习，自己改善内容的GPT。短短几个周的时间，其GitHub星星已经超过76.8K，超过了pytorch的gitHub星星数量。

Auto-GPT

Auto-GPT是一个基于GPT-4，开源的可以自己学习的语言大模型，当然若想使用Auto-GPT，我们需要有Chat-GPT的API key，且模型基于GPT-4，若想体验比较强大的GPT功能，我们需要使用付费账号进行体验。既然Auto-GPT是一个可以自主学习的模型，是否我们没有Chat-GPT的API key，它是否可以使用互联网上的资源进行问答呢？答案是肯定的，虽然没有Chat-GPT的加持，Auto-GPT依然可以爬取互联网资料进行一些问答回复。

Auto-GPT需要使用付费账号

Auto-GPT代码实现

既然是开源的代码，我们就分享下如何来使用Auto-GPT，首先Auto-GPT基于ChatGPT开发，我们需要有一个API key，当然也不是必须的。然后python代码版本大于3.8.0。

https://github.com/Significant-Gravitas/Auto-GPT 项目地址

本代码运行在Google colab上，若运行在本地，请把命令行的叹号去掉。

!git clone https://github.com/Significant-Gravitas/Auto-GPT.git
'''
Cloning into 'Auto-GPT'...
remote: Enumerating objects: 5702, done.
remote: Counting objects: 100% (2/2), done.
remote: Compressing objects: 100% (2/2), done.
remote: Total 5702 (delta 0), reused 0 (delta 0), pack-reused 5700
Receiving objects: 100% (5702/5702), 1.74 MiB | 10.29 MiB/s, done.
Resolving deltas: 100% (3828/3828), done.
'''

首先我们需要把Auto-GPT clone到Google colab本地，当然若运行到本地，也可以使用git命令clone到自己的电脑上面。若没有git命令，也可以直接到Auto-GPT的GitHub仓库，点击code>> download zip，下载整个项目工程到本地即可。

%cd Auto-GPT!pip install -r requirements.txt

然后，我们进入Auto-GPT文件夹下，使用pip命令安装需要的第三方库，等待安装完成后，就可以运行Auto-GPT了，本地也是同样的过程，在Auto-GPT文件夹下，CMD命令行，然后输入pip install -r requirements.txt来安装所有的库。

!Python -m autogpt

最后我们直接使用python -m autogpt来运行Auto-GPT。通过以上的步骤，我们就可以成功的搭建了自己的Auto-GPT，代码运行后，要求我们需要输入Auto-GPT的名字，角色，已经要完成的目标，若不输入，模型采用默认的名字角色等。设置完成后，就可以跟我们的Auto-GPT进行聊天了。

Warning: The file 'AutoGpt.json' does not exist.Local memory would not be saved to a file.
Welcome to Auto-GPT!  Enter the name of your AI and its role below. Entering nothing will load defaults.
Name your AI:  For example, 'Entrepreneur-GPT'
AI Name: 
Entrepreneur-GPT here!  I am at your service.
Describe your AI's role:  For example, 'an AI designed to autonomously develop and run businesses with the sole goal of increasing your net worth.'
Entrepreneur-GPT is: good
Enter up to 5 goals for your AI:  For example: Increase net worth, Grow Twitter Account, Develop and manage multiple businesses autonomously'
Enter nothing to load defaults, enter nothing when finished.
Goal 1: 
Goal 2: 

首先，这里输入了一个最简单的命令1+1=多少的问题，由于小编这里输入的是GPT-3的API-key，因此程序提示没有gpt-4模型，但是程序使用了GPT agent来响应了这个简单的计算。

Input:1+1=？
SYSTEM:  Human feedback: 1+1=？
 THOUGHTS:  I can use the 'evaluate_code' command to calculate 1+1.
REASONING:  The 'evaluate_code' command can be used to evaluate simple mathematical expressions.
PLAN: 
-  Use the 'evaluate_code' command to calculate 1+1.
CRITICISM:  I should have thought of using the 'evaluate_code' command earlier.
NEXT ACTION:  COMMAND = evaluate_code ARGUMENTS = {'code': '1+1'}
Enter 'y' to authorise command, 'y -N' to run N continuous commands, 'n' to exit program, or enter feedback for ...
Input:y
-=-=-=-=-=-=-= COMMAND AUTHORISED BY USER -=-=-=-=-=-=-= 
SYSTEM:  Command evaluate_code returned: Error: The model: `gpt-4` does not exist
 THOUGHTS:  I need to start a GPT agent to help me with this task.
REASONING:  Starting a GPT agent can help me with this task.
PLAN: 
-  Use the 'start_agent' command to start a GPT agent.
CRITICISM:  I should have thought of using a GPT agent earlier.
NEXT ACTION:  COMMAND = start_agent ARGUMENTS = {'name': 'math_agent', 'task': 'Calculate 1+1', 'prompt': 'What is 1+1?'}
Enter 'y' to authorise command, 'y -N' to run N continuous commands, 'n' to exit program, or enter feedback for ...
Input:y
-=-=-=-=-=-=-= COMMAND AUTHORISED BY USER -=-=-=-=-=-=-= 
SYSTEM:  Command start_agent returned: Agent math_agent created with key 0. First response: The sum of 1+1 is 2.
 THOUGHTS:  I can use the 'message_agent' command to get the answer to the question.
REASONING:  The 'message_agent' command can be used to get the answer from the GPT agent.

然后，我们再来一个稍微复杂的例子，让auto-GPT来写一个冒泡排序算法。这里使用了中文输入，但是模型能够识别，且重新翻译成了英文，然后，模型也给出了冒泡排序的算法代码，唯一不足的是模型输出的是文本，代码的缩进都没有了，而且模型搜索了Google，还提供了介绍排序算法的URL地址，真是强大了。感兴趣的小伙伴可以阅读一下下面的模型回复。

Input:使用python编写一个冒泡排序算法SYSTEM:  Human feedback: 使用python编写一个冒泡排序算法 THOUGHTS:  I can use the 'evaluate_code' command to evaluate the Python code for a bubble sort algorithm.REASONING:  The 'evaluate_code' command can be used to evaluate Python code.PLAN: -  Use the 'evaluate_code' command to evaluate the Python code for a bubble sort algorithm.CRITICISM:  I should have asked for more information about the bubble sort algorithm before attempting to evaluate the code.NEXT ACTION:  COMMAND = evaluate_code ARGUMENTS = {'code': '# Python program for implementation of Bubble Sort\n\narr = [64, 34, 25, 12, 22, 11, 90]\n\n# Traverse through all array elements\nfor i in range(len(arr)):\n\n # Last i elements are already in place\n for j in range(0, len(arr)-i-1):\n\n # traverse the array from 0 to n-i-1\n # Swap if the element found is greater\n # than the next element\n if arr[j] > arr[j+1] :\n arr[j], arr[j+1] = arr[j+1], arr[j]\n\nprint ("Sorted array is:")\nfor i in range(len(arr)):\n print ("%d" %arr[i]),'}Enter 'y' to authorise command, 'y -N' to run N continuous commands, 'n' to exit program, or enter feedback for ...Input:y-=-=-=-=-=-=-= COMMAND AUTHORISED BY USER -=-=-=-=-=-=-= SYSTEM:  Command evaluate_code returned: Error: The model: `gpt-4` does not exist THOUGHTS:  I can use the 'google' command to search for information on bubble sort algorithm in Python.REASONING:  The 'google' command can be used to search for information on the bubble sort algorithm in Python.PLAN: -  Use the 'google' command to search for information on the bubble sort algorithm in Python.CRITICISM:  I should have thought of using the 'google' command earlier.NEXT ACTION:  COMMAND = google ARGUMENTS = {'input': 'bubble sort algorithm in Python'}Enter 'y' to authorise command, 'y -N' to run N continuous commands, 'n' to exit program, or enter feedback for ...Input:y-=-=-=-=-=-=-= COMMAND AUTHORISED BY USER -=-=-=-=-=-=-=  SYSTEM:  Command google returned: b'[\n {\n "title": "Python Program for Bubble Sort - GeeksforGeeks",\n "href": "https://www.geeksforgeeks.org/python-program-for-bubble-sort/",  THOUGHTS:  I can use the 'google' command to search for information on bubble sort algorithm in Python. I can also use the 'write_to_file' command to save the code for the bubble sort algorithm in Python to a file.REASONING:  The 'google' command can be used to search for information on the bubble sort algorithm in Python. The 'write_to_file' command can be used to save the code for the bubble sort algorithm in Python to a file.PLAN: -  Use the 'google' command to search for information on the bubble sort algorithm in Python.-  Use the 'write_to_file' command to save the code for the bubble sort algorithm in Python to a file.

一些错误信息以及相关的log，可以使用如下命令行来查看

python -m autogpt --debug

当然，我们也可以使用语音输入

python -m autogpt --speak