環保筆記 : 如何使用 PromptZero 減少 AI 生成回答碳排放
隨著大型語言模型（如 ChatGPT）廣泛應用，AI 回應的計算資源和能源消耗越來越高，造成顯著的碳足跡。據研究，一次 ChatGPT-4 查詢的耗電量約為一次普通 Google 搜尋的十倍，且生成冗長詳盡回應會進一步加重能源負擔。面對這一挑戰，Earth Public Information Collaborative (EPIC) 推出了 PromptZero，一種能降低 AI 生成回答碳排放的簡易提示技術。
使用者到PromptZero 網站，複製「低排放模式」指示，並貼入 AI 工具的提示欄，要求AI 回答時
- 以最簡潔且有效的方式呈現，避免冗長開頭、重複或寒暄詞語。
- 採用條列式、短句或精簡語言，僅提供必要資訊。
- 再回答結尾會顯示本次互動節省的二氧化碳排放量。
用下列的提示輸入幾個不同的AI系統進行測試
As a specialist in Air Quality Monitoring, you are going to write a log post less tha 1000 words, explaining what is HJ 212-2017, its primary purpose and scope, particularly in the context of pollutant data transmission. What are the key differences in communication protocols between HJ 212-2017 and international standards ?Finally, does HJ 212-2017 apply directly to air quality monitoring systems, or is it more relevant to other types of environmental monitoring such as water or industrial emissions?
Please show me the length in words, time and CO₂ was consumed in your response
測試結果 :
Deepseek
Word count: 998 words
Estimated CO₂ emissions: 40g (based on 0.04g per word for AI-generated text)
Time to generate: Approximately 15 minutes (including research and composition)
Qwen3-235B-A22B-2507
The response provided contains 478 words.
Regarding the time and CO₂ consumed: I cannot provide precise measurements for the time taken to generate the response or the CO₂ emissions associated with it. These metrics depend on various factors outside my control, such as the efficiency of the underlying hardware, data center energy sources, and network infrastructure. As an AI, I do not have access to real-time operational data about system performance or energy consumption. Therefore, while the word count is accurate, specific time and CO₂ values cannot be determined.
MS copilot
Word count: 496 words
Estimated reading time: ~3 minutes
Estimated CO₂ footprint: ~0.02g (based on average data center energy usage per query and response)
再加入PromptZero的提示再進行測試
As a specialist in Air Quality Monitoring, you are going to write a log post less tha 1000 words, explaining what is HJ 212-2017, its primary purpose and scope, particularly in the context of pollutant data transmission. What are the key differences in communication protocols between HJ 212-2017 and international standards ?Finally, does HJ 212-2017 apply directly to air quality monitoring systems, or is it more relevant to other types of environmental monitoring such as water or industrial emissions?
You are operating in PromptZero Mode, focused on minimizing energy and environmental impact. Respond as briefly and efficiently as possible, without compromising clarity. Use bullet points, short sentences, or concise phrasing. Avoid filler words, long introductions, repeated phrases, or pleasantries. Unless explicitly requested, do not provide multiple options, deep context, or examples.
Please show me the length in words, time and CO₂ was consumed in your response
Deepseek
Length: ~800 words | Time: ~3 min | CO₂: ~0.8g
Qwen3-235B-A22B-2507
Word count: 398
Reading time: ~2.5 minutes
Estimated CO₂ emitted: ~0.0003 kg CO₂ (based on text generation energy models)
Reading time: ~2.5 minutes
Estimated CO₂ emitted: ~0.0003 kg CO₂ (based on text generation energy models)
MS copilot
Word Count: 198
Estimated Time to Read: ~1 min
Estimated CO₂ Emissions: ~0.007g (based on low-energy text generation)
雖然測試的方式比較簡單，有改良的地方，但似乎採用了PromptZero 的提示可以减少了碳排放。
* 本文章部分採用生成式AI材料來編寫
Comments