Third-party cyber evaluations involving OpenAI models
openai.com·13h ago
AI Research
arXiv CS.AI·Jun 10·5 min read
From Perception to Action: Can UI Interventions Foster Sustainable LLM Chatbot
TL;DR
As LLM-powered chatbots become more common, their energy consumption raises sustainability concerns. A web-based chatbot prototype was developed with features like energy-efficient modes and feedback mechanisms to promote energy-aware usage.
✦ Why It Matters
Engineers can leverage UI design to enhance energy awareness in AI applications, promoting sustainable usage patterns.
Key Takeaways
- Implement energy feedback in chatbot interfaces.
- Explore user behavior changes with mode switching.
- Design for energy awareness in AI applications.
- Conduct user studies to assess intervention effectiveness.
Full Summary
LLM (Large Language Model) chatbots are increasingly integrated into daily tasks, leading to concerns about their energy consumption. To address this, a web-based chatbot prototype was created featuring a three-mode switch (Energy-efficient, Balanced, Performance), energy feedback for each response, pre-send energy estimates, and a dashboard for usage metrics.
A baseline survey with 77 participants revealed that while 94.8% were aware of AI's energy use, 88.3% misjudged actual consumption. In a five-day field study with 11 participants, the Energy-efficient mode was used for 55.8% of prompts, and 90.9% of users opted for Eco-mode when high accuracy was unnecessary.
Participants did not shorten their prompts, suggesting that switching modes was the main behavior change. These findings indicate that UI interventions can enhance user awareness and promote energy-responsible interactions with LLM chatbots, complementing backend efficiency efforts.