К основному содержимому
MMentorHub
Начать

Does a transformer think? On Searle, Dennett, and the polite refusal to answer.

Глава 3 из 9
Идёт урок
TX…most teams reach for fine-tuning first. It feels rigorous.
Does a transformer think? On Searle, Dennett, and the polite refusal to answer.
02:48
14:22
Конспект урока

Searle's argument is annoyingly hard to dismiss and annoyingly hard to accept. He says: a person in a room, shuffling Chinese characters by rule, doesn't understand Chinese — and a computer doing the same thing doesn't either. Most people read it once, decide it's obviously right or obviously wrong, and move on. That's a mistake.

The argument, fairly

Imagine you don't speak Chinese. You sit in a room. Slips of paper come in with Chinese characters on them. You consult a giant rulebook in English: "if you see X, write Y." You write the response, slide it out. From outside, the room seems fluent. From inside, you understand nothing.

# Searle's structure, in pseudocode
def room(input_slip):
    rule = lookup(input_slip, rulebook)
    return rule.apply(input_slip)
# Question: is there understanding
# anywhere in this stack?

The most interesting rebuttal

It isn't from Dennett — though his is sharp. It's the Systems Reply: maybe you don't understand Chinese, but the room as a whole does. You're just one component. Searle finds this absurd. I find it the strongest move in the literature, and the conversation we'll have for the next 11 minutes.

Практика · 3 вопроса

Три небольшие задачи.

01

Reconstruct Searle's argument in three sentences without using the word 'understand'.

Подсказка · Try 'has semantics' or 'grasps meaning'.

02

Imagine the rulebook is replaced by a transformer's weights. Does anything in Searle's argument change?

Подсказка · Searle would say no. Why might Dennett disagree?

03

Is the Systems Reply a philosophical claim or an empirical one? Defend.

Подсказка · What evidence would settle it?

Собрано вISKRA.WORKSКейсХочу такой же →