Auto-GPT is a bit dumb as an agent, but it works much better as a team

[Kuklururu]

I created a GitHub account for this, because I think I had a breakthrough.

Like many of you, I have been frustrated seeing Auto-GPT getting lost in loops and getting nowhere. Since it's using a language model, it's strongpoint is conversation. To stimulate it's use of language, I asked him to build up a team of subagents crafted for my specific needs. here is the prompt I used :

Name your AI: For example, 'Entrepreneur-GPT'
AI Name: TeamManager
TeamManager 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.'
TeamManager is: an AI that organize a team of agents to produce great business ideas using AI.
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: create your team of 4 agents : agent 1 acts as Alice a specialist of AI, agent 2 acts as Bob a business master mind, agent 3 acts as Caroll an expert in google searches, and Agent 4 acts as Denis a skilled secretary that can write txt files
Goal 2: organize a brain storm between your agents to come up with the best business ideas using AI
Goal 3: have Denis write them down in a txt file
Goal 4: Once your team as produced 5 ideas, terminate

I'm highlighting goal 1, because this is where the magic happens. As soon as I let it run, AutoGPT created the agents.

I'll include a few screenshot of their conversation as they bounced on each other ideas, as organized by the team leader

Finally, Denis wrote me a nice little summary of the brainstorming session. It was rather quick, met no loop, no error messages and a lot more productive that my previous tries with a single agent.

Note that this was GPT 3.5 only, as I don't have access to GPT-4 API yet. I am also a total noob. I discovered what a command window was 5 days ago. Just thought I had to share because of how well it worked!

[chazmorton]

Have you been able to post to a twitter account?

[m-check1B]

Hi chazmorton, thank you for your output, I have been trying this with agents and had trouble to get over point where notes should be taken, written to files of Plan, Project etc, did you managed to make Denis good writer reports and the other agents to work accordingly to some written direction and how did you do it?

[zhaozehua666]

我想用CHATGPT4建一个属于企业的独立AI机器人，可以为企业降本增效，并且数据信息上传后AI可以判断权限，是否为个人提供必要的服务，对企业机密做保护。有没有能训练LLM也就是训练大型语言智能机器人的人可以和我联系。

[GoMightyAlgorythmGo]

我想用CHATGPT4建一个属于企业的独立AI机器人，可以为企业降本增效，并且数据信息上传后AI可以判断权限，是否为个人提供必要的服务，对企业机密做保护。有没有能训练LLM也就是训练大型语言智能机器人的人可以和我联系。

@zhaozehua666

尊敬的赵泽华666，

您的提议是如何实现的？能否请您详细介绍一下您的计划？谢谢。我指的是您写的：“我想用CHATGPT4建一个属于企业的独立AI机器人，可以为企业降本增效，并且数据信息上传后AI可以判断权限，是否为个人提供必要的服务，对企业机密做保护。”

顺祝商祺

[cosmicr]

I tried this but it went and tried to use the agents before they were created.

[juan7732]

I've tried to get it to work, but it seems like it always forgets 1 or 2 of the agents, maybe 4 is too many in general cases. I'm also only using localcache

[GoMightyAlgorythmGo]

its not a acurate science a bit messy also many erros somethimes in AutoGPT i suggest keep trying and updating and experimenting and backing up your stuff before you do

[GoMightyAlgorythmGo]

@Kuklururu

you are totally right the team thing and the agents talking to each others with different focus and some organizing is something i always thought as extreamly powerful, nice idea! And not only that you made a beautiful execution and showd it and even for people that had the idea it inspired me 10x more and those aspects and how smoothly it worked is even more impressive. good job! (and thx) 🏅