AI Under the Hood: AI Doesn't Read Words, It Reads Tokens (Part 1)
Many of us use AI at work every day without knowing what happens after we hit Enter.
Welcome to AI Under the Hood, a new series with one mission: to explain, in plain English, how AI actually works for people who use it but don't build it.
A Harvard Business School study of 758 consultants found that on tasks AI is good at, like brainstorming and writing, people using it finished 12.2% more work, 25.1% faster. However, on a business problem that hinged on details buried in the data, the AI gave a confident wrong answer, and the people using it were 19 percentage points less likely to get it right.
Knowing what's under the hood, then, is how you tell the two apart.
LET’S START
When you type a sentence into an AI tool, it never sees your words. It sees tokens.
A token is a chunk of text, about three-quarters of a word on average. A 100-word email is roughly 130 tokens.
Picture a manager who pastes a 40-page contract into AI and asks one ten-word question. AI usage is measured in tokens, small pieces of words, and everything the AI reads counts, not just your question. So that ten-word question uses roughly 27,000 tokens. For a company paying per use, tokens are dollars. On a monthly plan, they eat into your usage limit. Ask three follow-ups in the same chat, and the contract is counted four times.
WHY THE SERIES
So, why am I writing this series? Because once I understood how AI works under the hood, my results improved fast. I wish I had learned it first, not last. I’ll release these posts often, so you can skip the trial and error and build your AI skills sooner.
NEXT
Overall, tokens are the currency of AI. The posts ahead show tokens in action.
If you found this helpful, give it a like and share it with someone who needs it.
Gen Z, wondering whether college still matters with AI? See my companion series, The Educated Edge.
Next up, Part 2: parameters, and why one of the world's largest AI models is free to download but nearly impossible to run.
#AIUnderTheHood #AI #AILiteracy #HowAIWorks
Welcome to AI Under the Hood, a new series with one mission: to explain, in plain English, how AI actually works for people who use it but don't build it.
A Harvard Business School study of 758 consultants found that on tasks AI is good at, like brainstorming and writing, people using it finished 12.2% more work, 25.1% faster. However, on a business problem that hinged on details buried in the data, the AI gave a confident wrong answer, and the people using it were 19 percentage points less likely to get it right.
Knowing what's under the hood, then, is how you tell the two apart.
LET’S START
When you type a sentence into an AI tool, it never sees your words. It sees tokens.
A token is a chunk of text, about three-quarters of a word on average. A 100-word email is roughly 130 tokens.
Picture a manager who pastes a 40-page contract into AI and asks one ten-word question. AI usage is measured in tokens, small pieces of words, and everything the AI reads counts, not just your question. So that ten-word question uses roughly 27,000 tokens. For a company paying per use, tokens are dollars. On a monthly plan, they eat into your usage limit. Ask three follow-ups in the same chat, and the contract is counted four times.
WHY THE SERIES
So, why am I writing this series? Because once I understood how AI works under the hood, my results improved fast. I wish I had learned it first, not last. I’ll release these posts often, so you can skip the trial and error and build your AI skills sooner.
NEXT
Overall, tokens are the currency of AI. The posts ahead show tokens in action.
If you found this helpful, give it a like and share it with someone who needs it.
Gen Z, wondering whether college still matters with AI? See my companion series, The Educated Edge.
Next up, Part 2: parameters, and why one of the world's largest AI models is free to download but nearly impossible to run.
#AIUnderTheHood #AI #AILiteracy #HowAIWorks