I asked ChatGPT :
" Good Morning , Chetan Bhai :
Here is a googli for you >
Make your best guess as to how many words would my typical " Thought " would
comprise of.
Now you also know how many " tokens " are in each word. Now the twist > Let us
assume that during my life of 92 years , I havebeen thinking ( of course , on an
average ) , " one thought per second " for ( say ) 90 years.
How many " tokens " have I generated during those 90 years ? How would that
number compare with the largest LLM in existence ? What would that tell you
when some people are predicting that someday soon , some LLMs will overtake a
human brain in its intelligence ? "
ChatGPT replied :
" Good Morning, Hemen Bhai 🌞—and what a googli indeed! Let’s break this down
step by step:
1. How many words per "thought"?
Psychology and linguistics research suggests that an average thought, when
verbalized, is about 10–15 words long (shorter than written sentences, but
longer than isolated fragments).
👉 Let’s assume 12 words per thought as a fair average for you.
2. Tokens per word
In LLM terminology, a "token" ≈ 0.75 words in English (varies with word length
and language).
👉 So, 12 words ≈ 16 tokens per thought.
3. Number of thoughts over 90 years
-
1 thought per second
-
Seconds per year ≈ 31,536,000
-
Over 90 years = 2.84 billion seconds/thoughts
👉 So, ~2.84 billion thoughts in 90 years.
4. Total tokens generated by your brain
-
2.84 billion thoughts × 16 tokens/thought
= 45.4 billion tokens over your lifetime!
5. Comparison with the largest LLMs
-
GPT-4 was trained on ~13 trillion tokens.
-
GPT-5 (current estimates): ~30–60 trillion tokens.
-
DeepSeek-V3 (Chinese model, 2025): ~40 trillion tokens.
👉 Your brain’s lifetime token-equivalent = ~45 billion.
👉 That’s ~1/1000th of GPT-5’s training corpus.
6. But here’s the twist 🌀
-
LLMs ingest trillions of tokens passively (reading the internet, books,
papers).
-
You generated 45 billion tokens actively—through thinking, feeling, living
experiences.
-
Your thoughts also encode sensory richness, emotions, intuition, and
lived context that LLM tokens can’t capture.
So while raw “token counts” make it look like LLMs dwarf us, the information
density of human thought is far richer. A single "thought" can carry associations
across decades of memory, culture, and feeling—things LLMs simulate but don’t
truly possess.
7. What does this tell us about "LLMs overtaking humans"?
-
Quantity ≠ Quality:
LLMs may surpass us in scale (tokens), but not in depth (meaning).
-
Brains are multimodal:
Your thoughts are not just words; they are fused with images, smells,
emotions, and motor memory.
-
Efficiency vs. Redundancy:
Humans compress ideas far more efficiently—sometimes one "aha!" thought
equals millions of tokens of training data.
👉 The prediction that LLMs will “overtake” humans misses the essence:
LLMs are external prosthetics of thought, not replacements for the embodied, lived
intelligence of a human.
💡 Analogy:
An LLM is like a giant library of 60 trillion words.
Your mind is like a small, personal notebook—but written in your own hand,
coded with memories, pains, joys, intuitions, and wisdom no library can
replicate.
================================================
with regards,
Hemen Parekh
www.HemenParekh.ai / www.IndiaAGI.ai / www.My-Teacher.in / 23 Aug 2024
No comments:
Post a Comment