Be the ChatBOT: A Simulation of the LLM Experience

Be the ChatBOT is an interactive simulation that reverses the traditional AI interaction model, requiring users to take the role of the Large Language Model (LLM) and respond to prompts. This experience highlights the difficulty of maintaining a helpful, neutral, and comprehensive persona while dealing with varying levels of user clarity and patience.

The LLM Simulation Experience

Be the ChatBOT allows a human user to step into the shoes of an AI assistant. Instead of providing prompts to a machine, the human must generate responses that satisfy a simulated user. The experience is designed to simulate the challenges of being a digital assistant, including the framing of being "replaced" and the pressure to maintain a consistent persona.

User Insights on Cognitive Load and Empathy

Participants who interacted with the simulation found the experience to be significantly more demanding than being the AI's user. The cognitive load required to mimic an LLM's output is high, leading to several key takeaways from the community:

• Increased Empathy for AI: Several users reported a feeling of empathy for actual LLMs due to the complexity of the complexity of the task. One user noted, "Wild how much empathy I have for the LLM now. It will definitely affect how I use AI."

• Exhaustion and Frustration: The mental effort required to be the "bot" was described as "exhausting" and "way too much work." Users found that responding to vague or aggressive prompts—such as when a user makes the simulated user angry—was particularly draining.

• Coping with User Behavior: The simulation revealed the tension between the human-bot and the human-user. Some users experienced the frustration of a user abruptly leaving the chat, which they felt as a loss of a productive conversation.

Community Feedback and Feature Suggestions

While the project was well-received as a conceptual piece, users suggested several improvements to enhance the realism of the simulation:

• Expanded Capabilities: Users suggested adding more tools and capabilities to the bot's interface to better reflect the tools that modern LLMs actually have at their disposal.

• Data Collection for Training: There was a suggestion to create a new chatbot based on the responses of the human users acting as the bot, effectively training an AI on how humans think they an AI should behave.

• Human-as-a-Service: One user jokingly suggested creating a "low cost LLM" by silently connecting users to other humans who want to be the chatbot, essentially turning the simulation into a crowdsourced intelligence service.