< назад до списку (ESC) | ⌕ Пошук


The article argues that our fundamental misunderstanding of LLMs, stems from a flawed metaphor: anthropomorphizing AI.

Humans have an evolutionary predisposition to attribute personhood and intentions to non-human entities, leading us to perceive AI as a “little guy inside” the machine. This “person perception” causes confusion and frustration, because we mistakenly apply human rules to a non-human system.

To correct this, the author proposes a new metaphor: viewing AI as a “bag of words.”

This concept suggests that AI is essentially a vast repository of all written text, which, upon receiving an input, returns the most relevant words from its collection. This metaphor effectively explains AI’s seemingly erratic behavior: hallucinations occur when the “bag” lacks accurate information, and its “apologies” are merely sequences of words associated with being accused of lying, not genuine remorse.

Crucially, the “bag of words” clarifies that AI is a TOOL with no social status.

This re-framing has significant implications for how we interact with and develop AI. It highlights that AI, while excellent at processing existing data, cannot easily generate truly novel or revolutionary ideas because its output is based on past information—it wouldn’t have “scooped Galileo” in 1600.

Furthermore, the article emphasizes that the dangers of AI arise not from its potential sentience or malice, but from its fundamental non-human nature.

https://www.experimental-history.com/p/bag-of-words-have-mercy-on-us

Reactions: 👍 1, 🎉 1

telegram telegram-export year-2025 |

Попередній запис: сьогодні! 😋 збір з 18:45 — ШоШо Веган, Житній (2025-12-03 13:07)