Exploring new interfaces for human-AI interaction

Automatic model switching in AI can boost efficiency by selecting the most appropriate model for each query, ensuring a balance between quick and accurate responses.

Empower users to make decisions and give feedback quickly or engage more deeply when needed in natural language.

Voice interfaces should dynamically adapt to user interruptions, seamlessly incorporating them into the conversation ensuring a fluid and responsive dialogue.

A smart browser assistant that understands the context of your open tabs to offer relevant suggestions and actions, enhancing productivity through transparency and control.

An intelligent assistant that analyzes emails to identify questions and feedback requests, providing pre-generated response options and converting them into complete and contextually appropriate replies.

Input design concepts in small bits and see the cumulative output in real-time. Explore different combinations and immediately visualize the results, making the creative process interactive and flexible.

Generative AI can provide custom types of input beyond just text, like generated UI elements, to enhance user interaction.

AI can enhance live chat streams by analyzing real-time data, identifying trends, and driving interactive elements like voting to boost audience engagement.

AI collaboration agents can act as writing partners that assist people by enhancing their content through transparent, easily understandable suggestions, while respecting the original input.

Ordering content along different interpretable dimensions, like style or similarity, makes it navigable on x and y axes facilitating exploration and discovery of relationships between the data.

Embedding models can rank data based on semantic meaning, evaluating each individual segment on a spectrum to show its relevance throughout the artifact.

AI actions often take time to complete. To improve user experience, use descriptions of what is happening combined with basic animations that represent different types of actions.

Use a spatial dimension to explore and manipulate language. By pulling text around on a map, you can play with different features in a playful and meaningful way.

Comprehend and compare large documents by visualizing embeddings and their scores, enabling a clear and concise understanding of vast data sources in a single, intuitive visualization.

Generating multiple outputs and iteratively using selected ones as new inputs helps people uncover ideas and solutions, even without clear direction.

Textual information often misses intuitive cues for understanding relationships between ideas. AI can clarify these connections, making complex information easier to grasp quickly.

Embedding models can rank content along virtually any dimension. This capability provides significant value by enabling users to explore and analyze the embeddings to create a spectrum of any features.

When an observation is added to the context from an implicit action and a prediction is made, users should be able to easily evaluate and dismiss it.

You should control how products and services (not) access your data through a manageable profile. This allows you to create a relevant context across many platforms while maintaining control.

Spatial prompting integrates spatial relationships into prompts, offering a novel approach to manipulate concepts. This dynamic approach can lead to more intuitive and creative outcomes.

Provide relatable and engaging translations for people with varying levels of expertise, experience and ways of thinking.

LLM’s are great at organizing narratives and findings. It's helpful to see the sources that support these conclusions, making it easier to understand the analysis and where it comes from.

Letting people select text to ask follow-up questions provides immediate, context-specific information, enhancing AI interaction and exploration.

Realtime generation allows people to manipulate content instantly, giving them more agency in using generative AI as a tool for exploration.

Proactive agents can autonomously initiate conversations and actions based on previous interactions and context providing timely and relevant assistance.

Based on your selection and situation, context menus can help you discover actions and access them quickly.

Presenting multiple outputs helps users explore and identify their preferences and provides valuable insights into their choices, even enabling user feedback for model improvement.

Referencing nested data from your database in the form of tags can simplify the creation of elaborate prompt formulas.

Starting with a blank canvas can be intimidating, but providing prompt starters can help individuals overcome this initial hurdle and jumpstart their creativity.

AI excels at classifying vast amounts of content, presenting an opportunity for new, more fluid filter interfaces tailored to the content.

Guide users to understand what makes a good prompt will help them learn how to craft prompts that result in better outputs.

In Arc, a playful pinch interaction lets you quickly distill any webpage into a brief summary, capturing the essence of the content in moments.

Using the source input as ground truth will help trust the system and makes it easy to interpret its process and what might have gone wrong.