bitforms gallery is pleased to announce FOREVER MEETINGS, Sarah Rothberg’s second exhibition with the gallery and debut New York solo exhibition. FOREVER MEETINGS continues the artist’s research on the dynamics of conversation and large language models, culminating in a new body of generative artwork from which the exhibition takes its name.

Within each work a custom software program titled FOREVER MEETINGS prompts two AI agents, embodied as avatars, to have infinite, never-repeating conversations. The program uses a set of topics, conversation styles, and pairs of opposing personas as system prompts within each work. For example, one agent might be instructed to “role play as someone very selfish”, while the other takes on the persona of selflessness. As the program exchanges responses, the discourse evolves and the generated script animates the avatars. All prompts are authored by the artist, representing dichotomies in the artist's own life. Their subjects engage ideas of morality, ownership, power, and relationships. FOREVER MEETINGS allow both the artist and the audience to observe the tendencies and patterns of AI models. 

Three works using the FOREVER MEETINGS software are framed by bespoke hand-sculpted forms. These pieces introduce an adapted version of Rothberg’s software where two set personas cycle through one question per day and one conversation type per hour. FOREVER MEETINGS: DIALECTICS, BABY! focuses on questions of gender, feminism, intergenerationally, and the nuclear family. While this theme never changes, every hour the avatars are programmed to shift their conversation type, ranging from argumentative to flirtatious language. In the work, FOREVER METINGS: STRANGE ALIGNMENTS two avatars pull and push against a wall towards each other, discussing the tradeoff of new technologies.This continues in perpetuity through a language model stored locally on the artwork’s harddrive, serving as a 2025 time capsule of conversational AI. FOREVER MEETINGS: GREY ZONE / OF TWO MINDS delves into the personal relationship to systems while PETTY TROLLEY PROBLEM discusses power and its expressions. Rothberg’s handmade frames are guided by intuition and mask the monitors beneath. As standalone sculptural objects, their glittering surfaces are charming, despite their spikes and hidden features. 

This juxtaposition is mirrored by Rothberg’s seemingly-emotive avatars that mask the work’s purely computational AI output. When witnessing two avatars in conversation, moments of profound brilliance can strike. At times, the avatars can imitate thoughtful moments of reflecting "when the right time to have a baby might be", or if "tourism is bad". However, Rothberg seeks to remind viewers that an AI chatbot’s “knowledge” is only in relation to the words that have been used to train the model. When the relatable statements are spoken by the avatars, it’s only because those same words exist in relation to each other in the training set enough times for AI to make that association. 

In 2023, bitforms presented Rothberg’s first exhibition titled SUPERPROMPT that explored how “personas” can be assigned to AI chatbots. At the time, this simple mechanism for shaping a language model's output were sometimes referenced as “superprompts”, and are now called “system prompts.” Sophie For You TTS (text-to-speech) is a web-based work that allows visitors to interface with an embodied interactive chatbot named Sophie. The character Sophie was born from a ChatGPT description of an AI-hallucinated, non-existent performance allegedly staged by Rothberg. Within this “performance” a character named Sophie surveyed the intersection of artificial intelligence and human emotion. In the spirit of this AI hallucination, Sophie was designed as the avatar for the artist’s conversational AI interface. When chatting with Sophie online, users can select a superprompt as a set of speaking conditions. The connection to a cloud-based API for the LLM and text to speech in SOPHIE FOR YOU means that this work changes as the technology evolves, unlike the static FOREVER MEETINGS which are tied to local models.  

HumanTTS: YOU (human text-to-speech) grants the audience its first gallery encounter with Rothberg’s FOREVER MEETINGS software, offering viewers a chance to read and recite an AI-generated text. The text, presented like a teleprompter for a conversation between two people, is situated on screens outfitted with microphones. Two monitors stand behind the microphones. As text appears, viewers can activate the work by speaking aloud into the microphones. Their voice animates the avatars in real-time. The script advances in an infinite, never-repeating loop that is echoed throughout the gallery between machines, avatars, and the artist. 

Across the gallery, HumanTTS: ME (me text-to-speech) adds humanness back into the generative scripts produced by FOREVER MEETINGS. In this video, the artist embodies two avatars while she orates scripts from her software. While YouTTS invites user participation, MeTTS is a curated focus on discussions created by the software then read by the artist. Both text-to-speech pieces invite the viewer to think deeply about their own language—how it is used to inform large language models and how these models enfold back into our language.Hypothetical answers to Rothberg’s own questions of virtue and integrity, such as “Is greed natural?", “Is making art a type of sickness”, and “Why do people hold hands??” are read aloud. The artist’s own voice echoes in cacophony alongside visitors, computer-generated AI transcripts, and Sophie’s recitations.  

FOREVER MEETINGS is supported by Onassis ONX.

The artist would like to graciously thank The Jerome Foundation for their support. The exhibition is produced in collaboration with Yotam Mann (AI System Engineer), Marpi (Animation System and Unity Developer), Nailah Hunter (Original Soundscape), Claire Hentschker (Frame Sparkle Engineer), and Han Zhang (Production Assistant).

See additional credits here.