HOW AI ACTUALLY WORKS AND FAILS
HOW AI ACTUALLY WORKS AND FAILS
TOKENS
AI reads in chunks called tokens, roughly ¾ of an English word each. It doesn’t read words. It predicts the next chunk.
HALLUCINATION
AI states made-up facts with total confidence. Humans do too: our memories fill gaps on the fly. Always ask for sources.
THE ELIZA EFFECT
In 1966 a simple chatbot called ELIZA mirrored people’s words back. Users poured their hearts out to it. We see minds everywhere.
MORAVEC’S PARADOX
The hard stuff (chess, calculus) is easy for machines. The easy stuff (folding laundry, reading a room) is hard. Your human skills are the moat.
THE JAGGED FRONTIER
In a 2023 study of 758 BCG consultants, AI made them faster and better on some tasks and worse on others, without any obvious warning about which was which.
MOVE 37 + KASPAROV’S LAW
In 2016 AlphaGo played a move experts gave 1-in- 10,000 odds. Kasparov’s lesson from chess: a human with a machine and a good process beats either alone.
