After creating ChatGPT Assistant, it must also be properly trained so that it can advise users as clearly and competently as possible based on the available information.
Files with the following extensions can be uploaded: c, cpp, cs, css, doc, docx, go, html, java, js, json, md, pdf, php, pptx, py, rb, sh, tex, ts, txt.
The file size must not exceed 40 MB. There is also a limit on the number of OpenAI tokens; the data in the file cannot exceed 5,000,000 tokens.
You can calculate the approximate number of tokens here using OpenAI's tokenizer.
The assistant only works with text – any other data types, such as images, will be ignored.
Tables cannot be used in the file (not even Markdown tables). ChatGPT does not read tables the same way humans do and may read them incorrectly and distort the data.
It is not recommended to upload duplicates of the same file and duplicate the information in them. The presence of duplicate data can negatively affect the accuracy of responses and increase the cost of the request.
The file content should be as concise and structured as possible: it is recommended to build a structure of headings and subheadings and divide the information into points where possible. The lack of structure in the data can also negatively affect the accuracy of responses and the cost of requests.
There is no need to write instructions for the bot in the file — they are set separately in the Instructions tab of the assistant editor.
It is recommended to divide the data into several files according to their purpose. For example, one file can contain information about the company, another — frequently asked questions, and the next — information about the company's products.
It is recommended to format similar information as a single item in the file, if possible. If this is not possible, similar data should be located as close to each other as possible within the file.
The content of the file directly affects the quality of the response and the final cost of each request.
An example using an assistant with an incorrect and correct structure:
Therefore, it is strongly recommended to train assistants properly to prevent such cases.