-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathhow-it-works.html
61 lines (56 loc) · 2.28 KB
/
how-it-works.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
<h1>How does it work?</h1>
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.
<h2>Core Implementation</h2>
<h3>Token Counting</h3>
<li>Character-based token estimation</li>
<li> Real-time conversation tracking</li>
<li> Basic approximation (configurable chars/token ratio)</li>
<h3>Emissions Calculation</h3>
Total Energy = (Input_Tokens * Input_Factor + Output_Tokens * Output_Factor) * PUE
<br/>
Total Emissions = Total Energy * Grid_Factor
<p>
<h3>Configurable Parameters</h3>
<b>1. Infrastructure</b>
<li>PUE (datacenter <a href="https://en.wikipedia.org/wiki/Power_usage_effectiveness">Power Usage Effectiveness)</a></li>
<li>Default: 1.125 - Range 1.0-2.0</li>
<br/>
<b>2. Energy Grid</b>
<li>Grid Emissions Factor</li>
<li>Default: 383 gCO2e/kWh</li>
<li>Local: <a href="https://app.electricitymaps.com/map/72h">ElectricityMap</a></li>
<br/>
<b>3. Token Estimation</b>
<li>Characters per Token</li>
<li>Default: 4 chars/token - range (3-5)</li>
<br/>
<b>4. Energy Factors</b>
<br/>
<li>Input Token Energy (For context processing): Default: 0.002 Wh/token</li>
<li>Output Token Energy (For generation processing): Default: 0.01 Wh/token</li>
<p>
<h2>Important Limitations</h2>
<h3>Estimation Accuracy</h3>
<li>Token count is based on character count approximation on web page using DOM method</li>
<li>Cannot detect cached responses</li>
<li>Simplified token energy estimations based on the <a href="https://huggingface.co/spaces/genai-impact/ecologits-calculator">ecologITs calculator</a></li>
<br/>
<h3>Scope Limitations</h3>
<li>Only tracks simple text interactions</li>
<li>Does not account for:</li>
<li>Complex reasoning tasks (agentic)</li>
<li>Multimodal processing</li>
<li>Local model deployment</li>
<li>Hardware lifecycle embodied carbon emissions</li>
<li>Data center embodied carbon emissions</li>
<h3>Technical Limitations</h3>
<li>Basic token estimation</li>
<li>Conservative energy calculations</li>
<li>Limited to ChatGPT web interface</li>
<li>Chrome browser only</li>
<h3>Browser Compatibility</h3>
<li>Chrome version tracking</li>
<li>Minimum version requirements</li>
<li>version compatibility logging</li>
<h3>ChatGPT Interface compatibility</h3>
<li>Version detection (logs)</li>