Interacting with an eco-feedback LLM interface surfaces gaps in students' understanding of digital sustainability and supports learning about the environmental impact of everyday LLM use¶

Claim · All claims
Evidence · 1 study · qualitative r3 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Participants reported learning through the interface, e.g., that even a simple chat has a carbon emission and that latency is an environmental factor. → Hyeonwook Kim 2026 q2 i? Qualitative analysis revealed a significant baseline gap in understanding GHG and latency as environmental factors, which the interface began to fill. → Hyeonwook Kim 2026

Evidence¶

Hyeonwook Kim 2026¶

Hyeonwook Kim, Xuesi Chen, Alex Cabral, Cindy Kaiying Lin, Udit Gupta, Josiah Hester. (2026). When LLMs Slow Down: How Environmental Impacts Mediate University Students' LLM Usage. https://arxiv.org/abs/2608.23968 doi:10.48550/arxiv.2608.23968

q2 · i? · qualitative · r3

Reflexive thematic analysis (Braun and Clarke's approach) of open-ended survey responses and group discussions produced 85 initial codes consolidated into 36 sub-themes and four themes. "Several participants explicitly reported learning through the interface", including that latency is an environmental factor.

"Several participants explicitly reported learning through the interface, noting that “even a simple chat has a carbon emission” and that it helped them understand “the trade off between being fast and being green.”"

Discussion¶

Related Claims¶

• Concrete, outcome-oriented eco-feedback visualizations with real-world analogies are rated more favorably than abstract metrics — related
• Recognizing carbon-saving impacts increases the likelihood of selecting eco modes in LLM interactions — related
• Most participants expressed reluctance to adopt an eco-feedback LLM for regular use, with nearly two-thirds reporting some degree of unwillingness — related
• Self-hosted open-weight LLMs offer privacy, reproducibility, and environmental advantages over commercial LLM interfaces in educational ITSs — related
• Control-group students identified the three sustainability pillars but confined self-reported application largely to environmental considerations — related
• The review reports that training a single deep learning model may have a carbon footprint comparable to several automobiles over their lifespan
• Eco-mode selection in an LLM interface declines sharply as response latency increases, dropping from about 45% of interactions in eco mode 1 to below 5% in eco mode 5 — related
• Stronger eco-mindedness raises initial eco-mode preference, but this advantage diminishes as response latency increases — related