免费的 GPT-4 Turbo 香不香？Coze 扣子海外版、中文版双发，深度评测 + 手把手教程 | 回到Axton

The video introduces Coze, an AI platform by ByteDance that allows users to create conversational bots using GPT-4 Turbo with up to 128K context, for free. The host, Axton, discusses the appeal of such a platform and outlines the agenda for the video, which includes testing Coze's basic functionalities, creating a dialog bot, exploring multi-agent chatbots, and discussing knowledge base creation. The platform is aimed at the overseas market but has recently launched a Chinese version, Coze.cn, which uses a different model called the 'Lark' model, making it unsuitable for those looking to use GPT-4 for free. The video proceeds to guide viewers through the process of registering on Coze.com, navigating its interface, and understanding its capabilities, such as creating bots, integrating plugins for extended functionality, and utilizing workflows and knowledge bases.

Axton demonstrates how to create a single-agent bot on Coze.com, emphasizing the model selection, including GPT-4 Turbo with 128K context. He explains the features available for bot creation, such as adding plugins like DALL.E 3 and GPT-4V, integrating workflows for complex tasks, and utilizing the bot's memory, which includes creating knowledge bases and setting fixed variables. The video also shows how to test the bot's functionality through a preview window, create a knowledge base by uploading a CSV file and fetching online data via URL, and set up variables for specific information that doesn't require AI inference. Axton also touches on the new database feature for structured information storage and the process of publishing the bot to social media platforms.

The host guides viewers on how to add a knowledge base to their bot by creating a new one named 'Content Collection' and populating it with units of information. He shows two methods of adding knowledge: through a URL to fetch online blog posts and by uploading a CSV file containing video titles and links. Axton also introduces the concept of variables for storing explicit information and the database feature, which allows for the addition of structured data like常用链接 (commonly used links). He demonstrates adding, querying, and manipulating records within the database using natural language commands and SQL statements, highlighting the efficiency and precision this brings to the bot's responses.

Axton explores the database feature further by showing how to add, query, and delete records using natural language inputs. He successfully adds three records to the 'URLS' database and verifies the addition through a query. The video also demonstrates the deletion of specific records and the ability to clear all records from the database. However, it is noted that the database appears empty when accessed through Discord, suggesting a potential discrepancy between the platform's interface and the actual database content.

The video showcases the creation of a multi-agent bot, where each agent functions as a separate AI with isolated conversations. Axton explains the concept of 'jump conditions' that dictate the transition between different agents based on specific input criteria. He demonstrates the setup of these conditions and the resulting dialogue flow between two agents named Tom and Jerry. The host also notes the limitations due to sparse documentation and advises waiting for official documentation for advanced features.

Axton summarizes the advantages of Coze, highlighting its ease of creating bots and the ability to use GPT-4 and GPT-Turbo for free. He discusses the potential for users to create their own plugins and the early stage of advanced functionalities like workflows. The host expresses curiosity about the platform's future direction, whether it will focus on AI or evolve into a full-fledged automation platform like Make or Zapier. He also mentions Coze's advantage in integrating with domestic social platforms, which is a shortcoming of other tools. The video concludes with an invitation to join AI Elite Academy for further learning in AI application and prompts engineering.