Implementation

Hardware

gpTea’s hardware consists of three parts, the cup, the tray, and the teapot. The cup features a 1.28 inch round display, ESP32 microcontroller, microphone, and 1 liPo battery. The body of the cup has a magnifying lens built into its geometry to blend the display into the liquid content once the cup is filled. The tray houses a speaker, a stepper motor that manipulates the teapot’s rotation via a timing belt, and a rotary encoder equipped with 2 haptic motors for user feedback and one hall effect sensor for detecting the presence of the cup. All the electronics housed in the tray are connected to an arduino nano microcontroller.

Feel free to reach out to my project partner, Kevin Tang, who executed all the hardware + fabrication!

Prompt Engineering

Adopting the teapot's original design and interactions

We mapped interactions between the user and AI through the original affordances of the teapot – just as the teapot pours tea into a teacup, gpTea augments that familiar and beloved moment by telling, or “pouring,” a story which comes to life at the bottom of the cup. With the cup of tea in hand, the user then drinks the tea just as the told story becomes a part of them. Pouring and drinking become metaphorically linked with listening and absorbing/consuming through the mapping of these interactions.

Designing for poetry over productivity

Though gpTea’s ability to speak and pour tea on its own make it feel “intelligent”, it is the interplay of tea and light which imbues it with poetic qualities. We could have easily displayed the generated the image to the user on an LED screen, but with the generated image appearing at the bottom of the teacup, the tea becomes a blurred boundary between the fictitious, historical, surreal and real. The tea functions then as the “transitional space” between internal fantasy and external reality, an area of psychological experience that engages the imagination.

The visual distortions and discoloring from the movement and hue of the tea create a “digital sfumato” that paints the imagined memory with uncertainty. Coincidentally, the image generated by DALL-E is also nebulous, abstract, and sometimes even eerie. These ambiguities result in an ambience. The imprecision of the display encourages users to fill in gaps in information and to ultimately think more deeply about the stories being told and about the people to whom the stories belong. The projection of light onto tea offers several effects: the rhythmic rippling and swirling motion of the tea creates an entrancing effect; the soft glow of light permeating through the murky tea is ethereal; the fleeting image provides an air of transience.

As the teapot pours autonomously, prompted by the user’s action to activate storytelling mode, the affordance of flowing tea, the rippling and reflective surface of the liquid, and the natural color of the tea saturates the displayed image, creating a natural patina.

Limiting touch points

Users interact primarily through the tea cup as the speaker, microphone, dial and display.

Like tea drinking, communicating through gpTea becomes dialogic -- one step must be executed before proceeding to the next, a continuous exchange occurs between user and object. Limiting touchpoints paces the user’s interaction with the object, creating a “slow” quality that invites reflection. We designed gpTea to be “slow” (as opposed to "fast technology" which demands performance efficiency over a limited time for a specific use-case) so that users can fully become intimate with one another by stopping to reflect at each moment and using the object over a long period of time. The ritualistic nature of the narrative delivery not only slows time down, but also closes space down into a cup – the main touch point.

The visual effects projected onto the cup, such as a blinking red light for recording, allow users to access the state of the LLO

Storing memories

To address concerns regarding privacy, we opted not to store the transcript of the user's conversation with the AI but rather have an LLM extract and summarize the user's story from the conversation before storing it. Through initial feedback from others, we realized that in doing so, we also provide the user an opportunity to leave a digital trace for the next user and introduce a tiny bit of discord into the system. The idea of a digital trace is fascinating as digital technologies are often programmed to refresh + reboot. It also sustains a thread of continuity between the ongoing conversation between the users through the LLO -- one user uses the LLO and leaves a digital trace that will influence the other user's experience and so on. It adds a delightful poetry to digital information, metaphorically akin to the residual tea leaves.

Chaining LLM prompts

We realized several benefits to this approach beyond improving the quality of generated outputs. For example, it naturally allows us to map each prompt with an interaction such that the user can focus on the experience of each interaction at hand and contributes to the “slowness” of the machine. The chaining of LLM prompts stretches time between each inputs-outputs exchange, allowing meaningful interaction to occur there. It also introduces the context from the user into the interaction experience, allowing users to “switch” or “exit” prompts. In gpTea, users are able to switch between a passive storytelling mode and an active storytelling mode depending on their desired interaction.