|
1 | | -<h1>How does it work?</h1> |
2 | | - |
3 | | -AI Wattch is a Chrome extension that estimates and displays carbon emissions from ChatGPT conversations. The focus is on education and awareness about AI energy consumption. |
4 | | - |
5 | | -<h2>Core Implementation</h2> |
6 | | -<h3>Token Counting</h3> |
7 | | -<li>Character-based token estimation</li> |
8 | | -<li> Real-time conversation tracking</li> |
9 | | -<li> Basic approximation (configurable chars/token ratio)</li> |
10 | | - |
11 | | -<h3>Emissions Calculation</h3> |
12 | | -Total Energy = (Input_Tokens * Input_Factor + Output_Tokens * Output_Factor) * PUE |
13 | | -<br/> |
14 | | -Total Emissions = Total Energy * Grid_Factor |
15 | | -<p> |
16 | | - |
17 | | -<h3>Configurable Parameters</h3> |
18 | | - |
19 | | -<b>1. Infrastructure</b> |
20 | | -<li>PUE (datacenter <a href="https://en.wikipedia.org/wiki/Power_usage_effectiveness">Power Usage Effectiveness)</a></li> |
21 | | -<li>Default: 1.125 - Range 1.0-2.0</li> |
22 | | -<br/> |
23 | | -<b>2. Energy Grid</b> |
24 | | -<li>Grid Emissions Factor</li> |
25 | | -<li>Default: 383 gCO2e/kWh</li> |
26 | | -<li>Local: <a href="https://app.electricitymaps.com/map/72h">ElectricityMap</a></li> |
27 | | -<br/> |
28 | | -<b>3. Token Estimation</b> |
29 | | -<li>Characters per Token</li> |
30 | | -<li>Default: 4 chars/token - range (3-5)</li> |
31 | | -<br/> |
32 | | -<b>4. Energy Factors</b> |
33 | | -<br/> |
34 | | -<li>Input Token Energy (For context processing): Default: 0.002 Wh/token</li> |
35 | | -<li>Output Token Energy (For generation processing): Default: 0.01 Wh/token</li> |
36 | | -<p> |
37 | | -<h2>Important Limitations</h2> |
38 | | -<h3>Estimation Accuracy</h3> |
39 | | -<li>Token count is based on character count approximation on web page using DOM method</li> |
40 | | -<li>Cannot detect cached responses</li> |
41 | | -<li>Simplified token energy estimations based on the <a href="https://huggingface.co/spaces/genai-impact/ecologits-calculator">ecologITs calculator</a></li> |
42 | | -<br/> |
43 | | -<h3>Scope Limitations</h3> |
44 | | -<li>Only tracks simple text interactions</li> |
45 | | -<li>Does not account for:</li> |
46 | | -<li>Complex reasoning tasks (agentic)</li> |
47 | | -<li>Multimodal processing</li> |
48 | | -<li>Local model deployment</li> |
49 | | -<li>Hardware lifecycle embodied carbon emissions</li> |
50 | | -<li>Data center embodied carbon emissions</li> |
51 | | -<h3>Technical Limitations</h3> |
52 | | -<li>Basic token estimation</li> |
53 | | -<li>Conservative energy calculations</li> |
54 | | -<li>Limited to ChatGPT web interface</li> |
55 | | -<li>Chrome browser only</li> |
56 | | -<h3>Browser Compatibility</h3> |
57 | | -<li>Chrome version tracking</li> |
58 | | -<li>Minimum version requirements</li> |
59 | | -<li>version compatibility logging</li> |
60 | | -<h3>ChatGPT Interface compatibility</h3> |
61 | | -<li>Version detection (logs)</li> |
| 1 | +<div class="container mx-auto p-6 max-w-4xl"> |
| 2 | + <h1 class="text-3xl font-bold mb-4">AIWattch: How does it work</h1> |
| 3 | + |
| 4 | + <p class="mb-6"> |
| 5 | + AI Wattch is a Chrome extension that estimates and displays carbon emissions from ChatGPT conversations. The focus is on education and awareness about AI energy consumption. |
| 6 | + </p> |
| 7 | + |
| 8 | + <h2 class="text-2xl font-bold mt-8 mb-4">Core Implementation</h2> |
| 9 | + |
| 10 | + <h3 class="text-xl font-semibold mt-6 mb-3">Token Counting</h3> |
| 11 | + <ul class="list-disc pl-6 mb-6"> |
| 12 | + <li>Character-based token estimation</li> |
| 13 | + <li>Real-time conversation tracking</li> |
| 14 | + <li>Basic approximation (configurable chars/token ratio)</li> |
| 15 | + </ul> |
| 16 | + |
| 17 | + <h3 class="text-xl font-semibold mt-6 mb-3">Emissions Calculation</h3> |
| 18 | + <p class="mb-4"> |
| 19 | + Total Energy = (Input_Tokens * Input_Factor + Output_Tokens * Output_Factor) * PUE<br> |
| 20 | + Total Emissions = Total Energy * Grid_Factor |
| 21 | + </p> |
| 22 | + |
| 23 | + <h3 class="text-xl font-semibold mt-6 mb-3">Configurable Parameters</h3> |
| 24 | + |
| 25 | + <div class="mb-6"> |
| 26 | + <h4 class="font-bold mb-2">1. Infrastructure</h4> |
| 27 | + <ul class="list-disc pl-6 mb-4"> |
| 28 | + <li>PUE (datacenter <a href="https://en.wikipedia.org/wiki/Power_usage_effectiveness" class="text-blue-600 hover:underline">Power Usage Effectiveness</a>)</li> |
| 29 | + <li>Default: 1.125 - Range 1.0-2.0</li> |
| 30 | + </ul> |
| 31 | + </div> |
| 32 | + |
| 33 | + <div class="mb-6"> |
| 34 | + <h4 class="font-bold mb-2">2. Energy Grid</h4> |
| 35 | + <ul class="list-disc pl-6 mb-4"> |
| 36 | + <li>Grid Emissions Factor</li> |
| 37 | + <li>Default: 383 gCO2e/kWh</li> |
| 38 | + <li>Local: <a href="https://app.electricitymaps.com/map/72h" class="text-blue-600 hover:underline">ElectricityMap</a></li> |
| 39 | + </ul> |
| 40 | + </div> |
| 41 | + |
| 42 | + <div class="mb-6"> |
| 43 | + <h4 class="font-bold mb-2">3. Token Estimation</h4> |
| 44 | + <ul class="list-disc pl-6 mb-4"> |
| 45 | + <li>Characters per Token</li> |
| 46 | + <li>Default: 4 chars/token - range (3-5)</li> |
| 47 | + </ul> |
| 48 | + </div> |
| 49 | + |
| 50 | + <div class="mb-6"> |
| 51 | + <h4 class="font-bold mb-2">4. Energy Factors</h4> |
| 52 | + <ul class="list-disc pl-6 mb-4"> |
| 53 | + <li>Input Token Energy (For context processing): Default: 0.002 Wh/token</li> |
| 54 | + <li>Output Token Energy (For generation processing): Default: 0.01 Wh/token</li> |
| 55 | + </ul> |
| 56 | + </div> |
| 57 | + |
| 58 | + <h2 class="text-2xl font-bold mt-8 mb-4">Important Limitations</h2> |
| 59 | + |
| 60 | + <h3 class="text-xl font-semibold mt-6 mb-3">Estimation Accuracy</h3> |
| 61 | + <ul class="list-disc pl-6 mb-6"> |
| 62 | + <li>Token count is based on character count approximation on web page using DOM method</li> |
| 63 | + <li>Cannot detect cached responses</li> |
| 64 | + <li>Simplified token energy estimations based on the <a href="https://huggingface.co/spaces/genai-impact/ecologits-calculator" class="text-blue-600 hover:underline">ecologITs calculator</a></li> |
| 65 | + </ul> |
| 66 | + |
| 67 | + <h3 class="text-xl font-semibold mt-6 mb-3">Scope Limitations</h3> |
| 68 | + <ul class="list-disc pl-6 mb-6"> |
| 69 | + <li>Only tracks simple text interactions</li> |
| 70 | + <li>Does not account for: |
| 71 | + <ul class="list-disc pl-6 mt-2"> |
| 72 | + <li>Complex reasoning tasks (agentic)</li> |
| 73 | + <li>Multimodal processing</li> |
| 74 | + <li>Local model deployment</li> |
| 75 | + <li>Hardware lifecycle embodied carbon emissions</li> |
| 76 | + <li>Data center embodied carbon emissions</li> |
| 77 | + </ul> |
| 78 | + </li> |
| 79 | + </ul> |
| 80 | + |
| 81 | + <h3 class="text-xl font-semibold mt-6 mb-3">Technical Limitations</h3> |
| 82 | + <ul class="list-disc pl-6 mb-6"> |
| 83 | + <li>Basic token estimation</li> |
| 84 | + <li>Conservative energy calculations</li> |
| 85 | + <li>Limited to ChatGPT web interface</li> |
| 86 | + <li>Chrome browser only</li> |
| 87 | + </ul> |
| 88 | + |
| 89 | + <h3 class="text-xl font-semibold mt-6 mb-3">Browser Compatibility</h3> |
| 90 | + <ul class="list-disc pl-6 mb-6"> |
| 91 | + <li>Chrome version tracking</li> |
| 92 | + <li>Minimum version requirements</li> |
| 93 | + <li>Version compatibility logging</li> |
| 94 | + </ul> |
| 95 | + |
| 96 | + <h3 class="text-xl font-semibold mt-6 mb-3">ChatGPT Interface Compatibility</h3> |
| 97 | + <ul class="list-disc pl-6 mb-6"> |
| 98 | + <li>Version detection (logs)</li> |
| 99 | + </ul> |
| 100 | +</div> |
0 commit comments