Synthesis inputs distill data from multiple sources into a single output. Synthesize inputs may include an element of summarization, but critically, it helps to restructure and organize the information first and foremost.

Within the interface, examples of this pattern include open canvas tools like Miro or Figjam, which allow the user to select multiple stickies or text inputs and have the AI synthesize the key themes, assigning categories and key words. Data analysis tools that summarize multiple data points or sets into a single chart or dashboard are another example of the pattern in use.

Midjourney's /blend function is another example applied to images, as it combines the most dominant tokens from two or more images into a single output.

Details and variations

This simple pattern makes use of other patterns like summarization and multi-modality to showcase its power

  • Let the user select what they want to combine, or let them put in an open prompt and do the initial work for them
  • Allow them the control to adjust the data to be synthesized. For example, Perplexity generates the initial sources for the user, but then allows the user to add or remove additional sources
  • Consider additional options as shown above in the Miro example. In this case, users can cluster by keyword, sentiment, or other qualities
  • Play with multi-modal outputs. A data table may be best synthesized into data visuals that can be easily parsed or included in a deck

Considerations

Positives

Ease of use
this input type requires no prompting by the user, and therefore can be used by anyone regardless of their skills with AI. They only need to select what they want to synthesize and then select the trigger, often a button.

Multi-modality

Synthesized outputs can take multiple forms depending on the users' need, making it highly adaptable. Data tables can be synthesized into visuals, multiple videos and articles can be synthesized into a single course targeted to the user, etc. Don't be afraid to push the boundaries to make the output more usable and useful for the person on the other side.

Concerns

Lack of caveats
When this is applied in a canvas setting, it may be difficult to warn users of the limitations of the AI in a way that breaks through. People could draw conclusions off of the synthesis that are not strong, or fail push back on the AI's organization.

Use when:
The user has a bunch of unstructured data and the AI can help organize it, or structured data that the AI can help pivot in a new direction.