Humans as AI Context – Ted Tschopp

<hr />

<h1 id="humans-as-ai-context">Human’s as AI Context</h1>

<hr />

<h2 id="words-speech-and-thought-rates">Words, Speech, and Thought Rates</h2>

<ul>
  <li>Humans speak ~15,000 words per day. <sup id="fnref:1"><a href="#fn:1" class="footnote" rel="footnote" role="doc-noteref">1</a></sup></li>
  <li>A person thinks between 4 and 30 words for every word spoken <sup id="fnref:2"><a href="#fn:2" class="footnote" rel="footnote" role="doc-noteref">2</a></sup></li>
  <li>A person thinks between 60,000 and 450,000 words per day. <sup id="fnref:2:1"><a href="#fn:2" class="footnote" rel="footnote" role="doc-noteref">2</a></sup></li>
</ul>

<h5 id="therefore">Therefore</h5>

<ul>
  <li>A person speaks at a rate of 125 to 175 words per minute <sup id="fnref:2:2"><a href="#fn:2" class="footnote" rel="footnote" role="doc-noteref">2</a></sup></li>
  <li>A person thinks at a rate of 1,000 to 3,000 words per minute</li>
</ul>

<hr />

<h2 id="human-words-per-token">Human words per token</h2>

<ul>
  <li>On average 1 word is approximately 1.33 tokens.</li>
  <li>Therefore, the average human speaks ~20,000 (19,950) tokens per day.</li>
  <li>Therefore, the average human thinks ~600,000 (598,500) tokens per day.</li>
</ul>

<hr />
<h2 id="current-llm-context-size">Current LLM Context Size</h2>

<ul>
  <li>GPT-1: 2,048 tokens (~1,500 words) - June 2018</li>
  <li>GPT-2: 2,048 tokens (~1,500 words) - February 2019</li>
  <li>GPT-3: 4,096 tokens (~3,000 words) - June 2020</li>
  <li>GPT-4: 8,192 tokens (~6,000 words) - March 2023</li>
  <li>GPT-4-turbo: 128,000 tokens (~96,000 words) - November 2023</li>
  <li>GPT-5-Standard: 128,000 tokens (~96,000 words) - July 2025</li>
  <li>GPT-5-Pro: 196,000 tokens (~147,000 words) - July 2025</li>
  <li>GPT-5-API: 400,000 tokens (~300,000 words) - July 2025</li>
</ul>

<hr />

<h2 id="human-speech-vs-human-thought-vs-llm-tokens-per-second-by-model">Human Speech vs. Human thought vs LLM Tokens Per Second by model</h2>

<ul>
  <li>Human Speech - ~2.08 tokens per second</li>
  <li>Human Thought - ~25 to 50 tokens per second</li>
  <li>GPT 1 - ~10 tokens per second</li>
  <li>GPT 2 - ~10 tokens per second</li>
  <li>GPT 3 - ~15 tokens per second</li>
  <li>GPT 4 - ~20 tokens per second</li>
  <li>GPT 4-turbo - 80 tokens per second</li>
  <li>GPT 5 - 50 tokens per second</li>
</ul>

<hr />

<h2 id="length-of-time-by-context-window">Length of time by Context Window</h2>

<table>
  <thead>
    <tr>
      <th>Model</th>
      <th>Context Size (tokens)</th>
      <th>Approx. Words</th>
      <th>Length of Human Speech</th>
      <th>Length of Human Thought</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>GPT-1</td>
      <td>2,048</td>
      <td>~1,500</td>
      <td>~0.075 days (1.8 hours)</td>
      <td>0.033 days (0.8 hours)</td>
    </tr>
    <tr>
      <td>GPT-2</td>
      <td>2,048</td>
      <td>~1,500</td>
      <td>~0.075 days (1.8 hours)</td>
      <td>0.033 days (0.8 hours)</td>
    </tr>
    <tr>
      <td>GPT-3</td>
      <td>4,096</td>
      <td>~3,000</td>
      <td>~0.15 days (3.6 hours)</td>
      <td>0.067 days (1.6 hours)</td>
    </tr>
    <tr>
      <td>GPT-4</td>
      <td>8,192</td>
      <td>~6,000</td>
      <td>~0.3 days (7.2 hours)</td>
      <td>0.133 days (3.2 hours)</td>
    </tr>
    <tr>
      <td>GPT-4-turbo</td>
      <td>128,000</td>
      <td>~96,000</td>
      <td>~4.8 days</td>
      <td>2.4 days (57.6 hours)</td>
    </tr>
    <tr>
      <td>GPT-5-Standard</td>
      <td>128,000</td>
      <td>~96,000</td>
      <td>~4.8 days</td>
      <td>2.4 days (57.6 hours)</td>
    </tr>
    <tr>
      <td>GPT-5-Pro</td>
      <td>196,000</td>
      <td>~147,000</td>
      <td>~7.35 days</td>
      <td>3.6 days (86.4 hours)</td>
    </tr>
    <tr>
      <td>GPT-5-API</td>
      <td>400,000</td>
      <td>~300,000</td>
      <td>~15 days</td>
      <td>7.2 days (172.8 hours)</td>
    </tr>
  </tbody>
</table>

<div class="footnotes" role="doc-endnotes">
  <ol>
    <li id="fn:1">
      <p><a href="https://www.scientificamerican.com/article/women-talk-more-than-men/#:~:text=In%20most%20of%20the%20samples,numbers%20to%20suit%20their%20tastes.%22">Some speak 4x more then others, while others speak a lot less</a> <a href="#fnref:1" class="reversefootnote" role="doc-backlink">↩</a></p>
    </li>
    <li id="fn:2">
      <p><a href="https://journals.sagepub.com/doi/abs/10.2466/pms.1990.71.3.1043">Given the need to account for the fact that this ratio isn’t a precise due to the non-linear nature of thought for each individual and the content being processed.</a> <a href="#fnref:2" class="reversefootnote" role="doc-backlink">↩</a> <a href="#fnref:2:1" class="reversefootnote" role="doc-backlink">↩<sup>2</sup></a> <a href="#fnref:2:2" class="reversefootnote" role="doc-backlink">↩<sup>3</sup></a></p>
    </li>
  </ol>
</div>