Our user study with creatives showed that Jigsaw can help them

• Discover new AI capabilities beyond well-known AI models
• Perform rapid and intuitive prototyping of complex AI-enabled workflows
• Document their creative process and explain it to others

(6/6)

Using the Assembly Assistant, the user can use natural language to describe a full workflow and Jigsaw will automatically recommend a chain of AI puzzle pieces.

(5/6)

Using Semantic Search, the user can use natural language to describe an AI capability and Jigsaw will recommend suitable AI puzzle pieces.

(4/6)

Sometimes models cannot perfectly align, such as cases that require reasoning. Jigsaw uses language models as “glue pieces.”

Glue pieces support translating text to prompts suitable for generative models, brainstorming ideas, or performing custom language-based reasoning.

(3/6)

We created 39 different AI model puzzle pieces that support modalities of text, image, video, 3D, audio, and sketches.

AI puzzle pieces are color-coded by their input and outputs and snap together when they are compatible.

(2/6)

🧩Introducing Jigsaw -- chain multimodal AI models together using AI puzzle pieces.

@chuanenlin will present the work at #CHI2024
🕒Monday 11am
📍AI and Interaction Design Session, 320 'Emalani Theater

🌐jigsaw.to
📄arxiv.org/pdf/2310.08574

(1/6)