Causal Map AI app

Using Causal Map AI

1. User Authentication

1. Access the Application: Open the Causal Map AI web interface in a browser.

2. Login:
• Click the "Log In" button in the navigation bar.
• Choose to log in with email/password or Google account.
• If you're a new user, you might need to register first.

3. Admin Check: The system automatically checks if you have admin privileges. At the moment, others are not allowed entry as admins have access to all files on CM3.

2. File Selection

1. View Available Files: Once logged in, you'll see a dropdown menu labeled "Select a file": all CM3 files.

2. Note that access is entirely read-only at the moment.

3. Choose a File:
• Click on the dropdown to see all available files.
• Select the file you want to analyze.
• Your most recent settings for sources, prompt etc for this file are loaded.

4. File Loading: The system loads the chosen file's statements table.

3. Source and Statement Selection

Source Selection

1. View Sources: After selecting a file, another dropdown appears showing "Select sources".

2. Choose Sources:
• This dropdown is multi-select, allowing you to choose multiple sources.

Statement Selection

1. View Statements: A third dropdown, showing "Select statements", becomes populated with the relevant statements. As in Causal Map 3, if you don’t select any statements, this implies all of the statements for the selected sources.

2. Choose Statements:
• Like sources, this is also multi-select.

3. Filtering: Coding operations only include selected (or implied) statements.

4. Clearing: there isn’t a Clear All button but if you click on the first of many sources or statements and press and hold your delete key, you can delete them all quickly

5. AI Processing Setup

1. Enter Prompts:
• In the "Autocode" tab, you'll find a text area.
• Enter your custom prompts here. Each prompt should be separated by '===='. You can have unlimited prompts.
• You can invite the AI to use ellipses […].
• You can invite the AI to provide an additional comment after the quote.
• You can invite the AI to number the links. This means that in subsequent passes you can see what has changed, e.g. some links have been left out or added. In subsequent passes you can ask the AI to add a comment about what has changed and why.
• Only ask for links! The app will not process A >> B >>C in one line.
• So a link might look like this:

1. A >> B //// verbatim quote, possibly with ellipses [//// optional comment]

• These prompts guide the AI in analyzing the text.

2. Adjust Chunk Size:
• Use the slider to set the chunk size (ranging from 500 to 50,000).
• This determines how the text is divided for processing. This algorithm is an improvement over CM3.
• Smaller chunks might be more precise but take longer (and are more expensive) to process.

3. Keep sources intact

• Alternatively, you can select Keep sources Intact which overrides the chunking and keeps each source as one chunk

1. Select AI Model:
• Choose from a dropdown of available models (e.g., gpt-4o, gpt-4o-mini).
• Different models have varying capabilities and processing speeds.
• We now include o1 and o1 mini, which may prove to be the best yet.
• However prices:
• o1 mini is similar to 4o, so try it!
• o1 is expensive, about FIVE TIMES the cost of 4o
• We also include pinned and dated versions. The dated versions should always provide the same results, whereas the pinned versions may point to newer models as they get incrementally updated.

6. Initiate Analysis

1. Start Processing: Click the "Process Documents" button.

2. Loading Indication: The system will show a loading or processing message.

7. Results Generation

• The system sends chunks of text along with your prompts to the selected AI model.

• Each chunk is processed separately, identifying causal relationships and other requested information.

• The prompt tokens and completion tokens usage is logged in the CM3 usage table which admins can view on the Account tab of CM3. Price is not calculated.

8. Results Display

1. Table Generation:
• Once processing is complete, a table appears in the "Output" tab.
• This table is interactive and contains the analyzed data.

2. Table Contents:
• Columns include: Prompt number, Link number, Cause, Effect, Quote, Comment, Statement IDs, and more.
• If your original output included link numbers, the Link number column reflects this, otherwise they are auto-numbered.
• Each row represents a causal relationship or insight identified by the AI.

9. Interact with Results

1. Sort and Filter:
• Click on column headers to sort the data.
• Use the filter inputs at the top of each column to narrow down results.
• In particular, if you want to look at the final result from the last prompt/pass, filter for its number in the Prompt Number column. So if you did three passes, filter for 3.
• The quotes are auto matched against the text, using a strict, not fuzzy matching algorithm - unlike CM3.

2. Review Insights:
• Read through the cause-effect relationships identified.
• Check the quoted text that supports each insight.

3. Cross-reference:
• Use the Statement IDs to link back to the original text if needed.

10. Further Analysis

1. Iterate Process:
• Based on initial results, you might want to adjust your prompts or selection.
• You can go back to steps 2-5 and run the analysis again with different parameters.

2. Export or Save:
• Coming soon.