FAQ: Answers to Frequently Asked Questions
General questions
Q1-1. What is generative AI?
A1-1. Generative AI is a generic term for AI (artificial intelligence) that can generate content such as text, images, and videos.
Q1-2. What does the add-in do?
A1-2. It is an add-in that links Excel and the Generate AI API, and can express requests and responses to the API as expressions.
Q1-3. Which generated AI is supported?
A1-3. It supports OpenAI's ChatGPT, Google's Gemini (former Bard/PaLM API), and Anthropic's Claude.
Q1-4. Is the add-in free?
A1-4. The add-in itself is free of charge. Generally, payment to the AI provider is required to use the generated AI.
Q1-5. Are the add-ins secure?
A1-5. The content entered into Excel is sent to each company's API endpoint via an encrypted communication channel. We do not store any text entered by the user.
Q1-6. What if I have other questions?
Questions about how to use functions
Q2-1. What is the BB.ASK function?
A2-1. It is a function to send one-question-and-one-answer type instructions to AI and make it answer. It is the most basic function and can be used for a variety of processing by devising prompts.
format
BB.ASK(prompt,[temperature generated],[max tokens],[model name])
Q2-2. What is the BB.CHAT function?
Q2-2. It is a function that sends a series of conversations to AI and makes it answer the rest. It enables you to do things in Excel that previously could only be done via API, such as separating role settings and prompts.
Format
BB.CHAT(role column,content column,[temperature generated],[max tokens],[model name])
Q2-3. What is the BB.CATEGORIZE function?
A2-3. It is a function that sends a text and a list of categories to AI and makes AI answer which category the text corresponds to. It uses function calling internally and is unlikely to generate unexpected responses.
format
BB.CATEGORIZE(text,classified list,[temperature generated],[max tokens],[model name])
Q2-4. What is BB.LOOKAT function?
A2-4. It is a function that sends an image link (URL) and instructions to be read from the image to AI and makes it respond.
*Only for ChatGPT
format
BB.LOOKAT(prompt,image link,[temperature generated])
Q2-5. What is the BB.MAJORITY function?
A2-5. It is a function that classifies text using one or more AI models and determines the response by majority vote. Specify the AI to be voted on in the model list argument, the classification in the classification list, and the majority voting method in the majority voting method argument.
Format
BB.MAJORITY(model list,text,classification list,[majority voting method],[generation temperature])
Majority method (default 1)
0 Returns a simple majority decision Returns #N/A if there is a tie in any of the cases
1 Return majority decision Return #N/A if there is no majority (default)
2 Return the decision of a two-thirds majority Otherwise, return #N/A
3 Return unanimous decision Otherwise return #N/A
Questions about settings
Q3-1. What is an API key?
A3-1. It is the data used to link the user to the payment account when using the GenerateAI API. It is also called a "private key," and must be kept secret from third parties.
Q3-2. Which model should I choose?
A3-2. For ChatGPT, it is better to use gpt-3.5 series for general tasks. For advanced tasks, the gpt-4 series models are more likely to provide appropriate answers, but at a higher cost.
Q3-3. What is "generating temperature"?
A3-3. It is the setting at which the generating AI generates the next token. The lower the temperature, the less likely it is to change, the higher the temperature, the more likely it is to change.
Q3-4. What is "maximum output length"?
A3-4. It is the maximum length of tokens output by the generating AI.
Q3-7. What is "suppress offensive generation"?
A3-7. This is a setting to check if there are any problems with the content prior to making a request to the API.
*Only for ChatGPT and Claude
Q3-8. What is "Calculation Mode"?
A3-9. Excel automatically recalculates when cell contents are changed or when a file is opened. On the other hand, requests to the API generally incur a cost. To reduce unexpected expenses, please set "Manual" when you want to suppress sheet recalculation.
Questions about fine tuning
Q4-1. What is "fine tuning"?
A4-1. Fine tuning is a method of customizing a standard AI model and adding additional data to obtain more desirable answers. Fine tuning requires at least 10 sets of question/answer combinations in the same format as the BB.CHAT function.
*ChatGPT only
| A | B | |
|---|---|---|
| 1 | system | You are a garbage collector at city hall. You deal with citizens in a friendly manner, but respond clearly to what you don't know and what you can't do with "I don't know" and "I can't do it. |
| 2 | user | Can aluminum foil pans be thrown out as garbage? |
| 3 | assistant | Aluminum foil pans for pot noodles, etc. are non-burnable trash. |
| ... | ... | ... |
| 10 | user | Is a milk carton combustible garbage? |
| 11 | assistant | Milk cartons that are white inside can be collected as recyclable waste, so please wash them, unfold them, tie them with string, and take them to the designated collection point. Milk cartons with silver-colored (aluminum-coated) insides are paper containers, so please take them to the designated collection point as burnable trash. Please take it to the designated collection point as burnable trash. |
Fine tuning can be expected to improve performance, but it is recommended to check the AI operation in advance by using the BB.CHAT function to illustrate questions and answers.
Q4-2. What is a "base model"?
A4-2. "Base model" is the trained data you want to customize for fine tuning. as of February 2024, we recommend using "gpt-3.5-turbo-1106".
Q4-3. What is "trained data"?
A4-2. "Training data" is the range of data you wish to send for fine tuning.
*Only for ChatGPT
| A | B | |
|---|---|---|
| 1 | system | You are a garbage collector at city hall. You deal with citizens in a friendly manner, but respond clearly to what you don't know and what you can't do with "I don't know" and "I can't do it. |
| 2 | user | Can aluminum foil pans be thrown out as garbage? |
| 3 | assistant | Aluminum foil pans for pot noodles, etc. are non-burnable trash. |
| ... | ... | ... |
| 10 | user | Is a milk carton combustible garbage? |
| 11 | assistant | Milk cartons that are white inside can be collected as recyclable waste, so please wash them, unfold them, tie them with string, and take them to the designated collection point. Milk cartons with silver-colored (aluminum-coated) insides are paper containers, so please take them to the designated collection point as burnable trash. Please take it to the designated collection point as burnable trash. |
In the above case, the range of training data is "A1:B11".
Q4-4. What is "suffix"?
A4-2. Suffix is a text to distinguish data sent for fine tuning. when OpenAI receives the data and the fine tuning is completed, the name with the specified suffix is determined.
*ChatGPT only
Q4-5. What is "integrity self-testing"?
A4-5. "Integrity self-diagnosis" is a function to check if the data to be sent for fine tuning meets the minimum requirements. Even if "normal" is indicated in the self-diagnosis, it does not necessarily mean that fine tuning is successful or that model performance is improved.
*ChatGPT only
Q4-6. What is "cost estimation"?
A4-6. This function is for estimating the cost of fine tuning. Fine tuning will cost you depending on the number of tokens at training and at input/output.
*ChatGPT only