It works by breaking a larger task into smaller sub-tasks and then spinning off independent Auto-GPT instances in order to work on them. The original instance acts as a kind of project manager, coordinating all of the work carried out and compiling it into a finished result. As well as using GPT-4 to construct sentences and prose based on the text it has studied, Auto-GPT is capable of browsing the internet and including information it finds there in its calculations and output. In this respect, it’s more similar to the new GPT-4 enabled version of Microsoft’s Bing search engine. It also has a better memory than ChatGPT, so it can construct and remember longer chains of commands.
ChaosGPT // LOL, JFF sci-fi story https://twitter.com/chaos_gpt
The script works by running an infinite loop that does the following steps:
1. Pulls the first task from the task list.
2. Sends the task to the execution agent, which uses OpenAI’s API to complete the task based on the context.
3. Enriches the result and stores it in Chroma/Weaviate.
4. Creates new tasks and reprioritizes the task list based on the objective and the result of the previous task. [link]
AgentGPT allows you to configure and deploy Autonomous AI agents. Name your own custom AI and have it embark on any goal imaginable. It will attempt to reach the goal by thinking of tasks to do, executing them, and learning from the results
There are many “agents”
The simple agent scenario/lifecycle:
1. Tasks analysis and execution plan generation (steps) // LLMs/AI-Models
2. Tasks steps execution (web search, data processing, services interaction, etc.) // AI-models, web-services, and even humans calls, API/scripts remote/local
3. Results analysis/summarization // LLMs/AI-Models
4. Are we finished? New/next tasks definition/analysis, scenario running.
Agent task example:
Write a story based on user defined setting and writer — editor pattern. Writer (LLM/human) writes episodes, editor (LLM/human) reads and gives improvement recommendations. Iterative process, each piece of text (episode) is rewritten many times until certain criteria are met (limited number of iterations the simplest one).
The future is already here, don’t be afraid if any such agent contacts you in the near future…
Experts are predicting that Auto-GPT, an AI tool that answers its own questions, could outshine ChatGPT. [post]
Auto-GPT Examples and Use-cases
Editor-Writer(s), Planner-Executor(s), Manager-Worker(s), etc. It’s the same as for people.
I see a lot of people saying “its a bad analysis! it sucks! who would use this” and I agree, its not very in depth. But you have to understand, I set this up in < 1 hour. It’s not optimized to do this. Someone will (maybe me) make better, more specific version of this agent soon. Its just the start