AutoGPT is unreliable, Microsoft launches an upgraded version! Editable autonomous planning process

Xi Xiaoyao Technology Talks Original
Author | iven

##AutoGPT## that is popular all over the Internet #[1]The number of collections on Github has exceeded 100,000. This self-planning and self-executing agent focuses on the self-adjustment and optimization within the artificial intelligence model for the first time. However, many netizens have found that the performance of AutoGPT is unstable, and infinite loops are the most common phenomenon. In addition, the execution speed of AutoGPT is very slow. According to tests by netizens, New Bing takes 8 seconds to complete the task, while AutoGPT took a full 8 minutes!

The way AutoGPT works makes it need to call the API many times for a single task. It is calculated that the cost of a single task exceeds 100 yuan! Obviously such costs are expensive for personal use.

Microsoft Research’s recent new work proposes Low-code LLM, which can collaborate with agents through drag-and-drop through simple visual operations.

This mode first allows GPT to generate a task flow chart, which is very similar to AutoGPT's self-planning and self-execution logic, but different Yes, users can intuitively and easily understand and modify the entire execution process, thereby effectively controlling the operation of artificial intelligence.

It is called "Low-code" because it adopts the concept of visual programming, and users can adjust the process with simple clicks and drags. For complex tasks, users can effectively control the agent with their own ideas or preferences.

Low-code LLM generates the flow chart in one conversation, and the cost of calling the API is basically negligible. Moreover, this one-time generation of the flow chart also avoids the problem of infinite loops in AutoGPT, making The service is more stable!

The author found that this work is placed in the Repo of Microsoft

TaskMatrix.ai

[2], which has exceeded 30k stars. ^{Visual ChatGPT}[3] is also from the same team. TaskMatrix.AI shows how to connect foundation models and a large number of APIs in various fields to implement Task Automation (Visual ChatGPT is a classic example in the visual field). The newly launched Low-code LLM can play a role in interacting with users, helping users to let AI better understand what the users want to do.

Paper address:

https://www.php.cn/link/de9240f5c623bf031dcf0fca9770db44

Paper title:

" Low-code LLM: Visual Programming over LLMs."

Open source code:

https://www.php.cn/link/141aa4fef48df77f954d60a373a3c322

Workflow

Planning LLM generates a structured flow chart for complex tasks, which is somewhat similar to AutoGPT. The idea of ​​​​self-planning according to the goals given by the user

1. Users modify the flow chart through defined low-code visual operations (including clicking, dragging, and text editing) to convey their preferences and opinions to LLM
2. Executing LLM executes commands according to the workflow modified by the user and generates answers
3. Users can refer to the current answers and continuously modify the flow chart until satisfactory results are obtained

Predefined 6 types of low code operations

The advantages of this mode are as follows:

1. More controllable generation results: users can directly understand and control the execution logic of artificial intelligence, making the results easier to predict and control, and more in line with user needs;
2. User-friendly interaction Interface: Users can intuitively see the execution process, and the click and drag method also makes the operation more convenient and improves work efficiency;
3. Wide application scenarios: This method can be applied to many fields, especially those for users The article proposes 4 typical cases in which your thoughts and preferences are crucial.

In addition, Low-code LLM can also be extended with external APIs to further enrich scenario applications. For example, efficiently convey user ideas and preferences and help users automate tasks. When connecting with other tools, various functions such as vision and voice can be integrated.

Both AutoGPT and Low-code LLM are working hard to improve the performance and effect of artificial intelligence models. The former focuses on self-optimization and learning within the model, and the latter focuses on the collaboration and interaction between users and models. These two methods can complement each other and achieve better performance in different scenarios and tasks.

The acknowledgment section of the paper also mentioned that part of this article was generated through this model of cooperation. It seems that in the future, it is no longer a dream for people and large models to work closely together to create.

The above is the detailed content of AutoGPT is unreliable, Microsoft launches an upgraded version! Editable autonomous planning process. For more information, please follow other related articles on the PHP Chinese website!