Fw: [心得] 蘋果揭露人工智能並不思考
※ [本文轉錄自 Stock 看板 #1eJk0_y9 ]
作者: huabandd (我是阿肥巴你頭) 看板: Stock
標題: [心得] 蘋果揭露人工智能並不思考
時間: Sun Jun 15 23:01:12 2025
https://is.gd/U7lWC8
看到一個影片，非新聞報導，僅創作者想法
討論蘋果揭露關於人工智能當中的最大謊言
之前也看過一個影片指出
AI無法產出指針在10:10以外的手錶
（網路手錶圖片通常指針停留在10:10）
自己嘗試過幾次確實是這樣
因為AI的訓練是基於現有大量資料
去產生一個與大量資料相似的產物給你
無法產生未曾有過資料的產物
（以上為個人猜測，若錯誤請指正）
雖然現在已有研究嘗試將AI接入人類神經元
未來思考的方式有可能因此改變
不再只是記憶，而是真正的推理
（但是這好像也是一件滿恐怖的事情？）
如果說給AI一個圍棋組，但不教他玩法
也不讓他參考棋類玩法，讓他自行摸索
就像嬰兒拿到玩具後自己亂玩那樣
AI能夠自行生成玩法是不是就算是成功？
中文翻譯
蘋果剛剛揭露了人工智慧中最大的謊言，即「思考」的假象，證明了像 Claude、Deepseek-
R1 和 OpenAI 的 o3-mini 等推理模型實際上並不會思考。蘋果沒有重複使用舊有的數學問
題，而是創造了這些模型從未見過全新謎題。結果令人震驚。
對於簡單問題，標準 AI 模型表現優於昂貴的推理模型，同時消耗更少的資源。對於中等複
雜度問題，推理模型顯示出一些優勢，但對於模擬現實世界情境的高複雜度問題，兩種類型
的模型都崩潰到 0% 的準確度。
這是確鑿的證據：蘋果給予 AI 模型逐步的演算法，就像給某人食譜一樣，但它們仍在相同
的門檻下失敗了。這證明它們無法遵循邏輯指令。該研究沒有發現任何形式推理的證據，得
出結論認為其行為最好用複雜的模式匹配來解釋。
簡單地更改名稱和問題會顯著降低性能。蘋果測試了河內塔（Tower of Hanoi）謎題，模型
在超過 100 步的移動中成功了，但在新穎的渡河謎題中僅僅 4 步後就失敗了。它們是記憶
模式，而不是進行邏輯推理。
隨著問題變得越來越難，「思考」模型使用的 token 越來越少，放棄得更快，而不是更深
入地思考。即使無限的計算資源也無濟於事。這揭示了數十億美元的 AI 投資是建立在錯誤
前提上的。
公司將複雜的模式匹配宣傳為真正的推理，誤導消費者對 AI 能力的認知。我們離通用人工
智慧（AGI）的距離並沒有大家想像的那麼近；我們正在處理的是非常複雜的自動完成系統
。當先進的 AI 推理模型連遵循簡單指令都失敗時，我們是否正生活在一個建立在幻想之上
的 AI 炒作泡沫中？
英文原文
Apple just exposed the biggest lie in AI with the illusion of "thinking," provin
g that reasoning models like Claude, Deepseek-R1, and OpenAI's o3-mini don't act
ually think. Instead of recycling math problems, Apple created brand new puzzles
these models had never seen. The results were shocking.
For simple problems, standard AI models outperformed expensive reasoning models
while using fewer resources. For medium complexity, reasoning models showed some
advantage, but for high complexity problems that mirror real-world scenarios, b
oth types collapsed to 0% accuracy.
Here's the damning evidence: Apple gave AI models step-by-step algorithms, like
giving someone a recipe, and they still failed at the same thresholds. This prov
es they can't follow logical instructions. The study found no evidence of formal
reasoning, concluding behavior is better explained by sophisticated pattern mat
ching.
Simply changing names and problems degraded performance dramatically. Apple test
ed Tower of Hanoi puzzles, where models succeeded with 100-plus moves but failed
at novel river crossing puzzles after just 4 moves. They memorized patterns rat
her than reasoned logically.
As problems got harder, "thinking" models used fewer tokens and gave up faster i
nstead of thinking deeper. Even unlimited computational resources couldn't help.
This exposes that billions in AI investment are based on false premises.
Companies market sophisticated pattern matching as genuine reasoning, misleading
consumers about AI capabilities. We are not as close to AGI as everyone thinks;
we're dealing with very sophisticated autocomplete systems. When advanced AI re
asoning models fail at following simple instructions, are we living in an AI hyp
e bubble built on illusions?
--
※ 發信站: 批踢踢實業坊(ptt.cc)
※ 轉錄者: Kai877 (27.240.233.101 臺灣), 06/16/2025 12:48:36
※ 編輯: Kai877 (27.240.233.101 臺灣), 06/16/2025 12:49:23
推
06/16 12:55,
10小時前
, 1F
06/16 12:55, 1F
→
06/16 12:57,
10小時前
, 2F
06/16 12:57, 2F
推
06/16 13:17,
9小時前
, 3F
06/16 13:17, 3F
推
06/16 13:30,
9小時前
, 4F
06/16 13:30, 4F
→
06/16 13:35,
9小時前
, 5F
06/16 13:35, 5F
推
06/16 13:39,
9小時前
, 6F
06/16 13:39, 6F
推
06/16 13:40,
9小時前
, 7F
06/16 13:40, 7F
推
06/16 13:41,
9小時前
, 8F
06/16 13:41, 8F
→
06/16 13:43,
9小時前
, 9F
06/16 13:43, 9F
→
06/16 13:46,
9小時前
, 10F
06/16 13:46, 10F
→
06/16 13:48,
9小時前
, 11F
06/16 13:48, 11F
→
06/16 13:54,
9小時前
, 12F
06/16 13:54, 12F
噓
06/16 14:05,
8小時前
, 13F
06/16 14:05, 13F
→
06/16 14:10,
8小時前
, 14F
06/16 14:10, 14F
→
06/16 14:10,
8小時前
, 15F
06/16 14:10, 15F
推
06/16 14:12,
8小時前
, 16F
06/16 14:12, 16F
推
06/16 14:25,
8小時前
, 17F
06/16 14:25, 17F
推
06/16 14:29,
8小時前
, 18F
06/16 14:29, 18F
→
06/16 14:30,
8小時前
, 19F
06/16 14:30, 19F
→
06/16 14:31,
8小時前
, 20F
06/16 14:31, 20F
噓
06/16 14:31,
8小時前
, 21F
06/16 14:31, 21F
→
06/16 14:32,
8小時前
, 22F
06/16 14:32, 22F
→
06/16 14:33,
8小時前
, 23F
06/16 14:33, 23F
噓
06/16 14:36,
8小時前
, 24F
06/16 14:36, 24F
→
06/16 15:04,
8小時前
, 25F
06/16 15:04, 25F
→
06/16 15:11,
7小時前
, 26F
06/16 15:11, 26F
推
06/16 15:20,
7小時前
, 27F
06/16 15:20, 27F
推
06/16 15:29,
7小時前
, 28F
06/16 15:29, 28F
推
06/16 15:42,
7小時前
, 29F
06/16 15:42, 29F
→
06/16 15:45,
7小時前
, 30F
06/16 15:45, 30F
→
06/16 16:10,
6小時前
, 31F
06/16 16:10, 31F
推
06/16 16:12,
6小時前
, 32F
06/16 16:12, 32F
→
06/16 16:12,
6小時前
, 33F
06/16 16:12, 33F
推
06/16 17:12,
5小時前
, 34F
06/16 17:12, 34F
噓
06/16 17:23,
5小時前
, 35F
06/16 17:23, 35F
噓
06/16 17:55,
5小時前
, 36F
06/16 17:55, 36F
推
06/16 17:58,
5小時前
, 37F
06/16 17:58, 37F
→
06/16 17:58,
5小時前
, 38F
06/16 17:58, 38F
推
06/16 18:11,
4小時前
, 39F
06/16 18:11, 39F
→
06/16 18:19,
4小時前
, 40F
06/16 18:19, 40F
推
06/16 18:30,
4小時前
, 41F
06/16 18:30, 41F
推
06/16 19:15,
3小時前
, 42F
06/16 19:15, 42F
推
06/16 19:30,
3小時前
, 43F
06/16 19:30, 43F
→
06/16 19:31,
3小時前
, 44F
06/16 19:31, 44F
推
06/16 19:35,
3小時前
, 45F
06/16 19:35, 45F
→
06/16 19:35,
3小時前
, 46F
06/16 19:35, 46F
噓
06/16 19:44,
3小時前
, 47F
06/16 19:44, 47F
→
06/16 19:44,
3小時前
, 48F
06/16 19:44, 48F
→
06/16 21:47,
1小時前
, 49F
06/16 21:47, 49F
→
06/16 21:47,
1小時前
, 50F
06/16 21:47, 50F
→
06/16 21:48,
1小時前
, 51F
06/16 21:48, 51F
推
06/16 21:53,
1小時前
, 52F
06/16 21:53, 52F
→
06/16 22:09,
55分鐘前
, 53F
06/16 22:09, 53F
噓
06/16 22:22,
42分鐘前
, 54F
06/16 22:22, 54F
→
06/16 22:39,
25分鐘前
, 55F
06/16 22:39, 55F
→
06/16 22:39,
25分鐘前
, 56F
06/16 22:39, 56F
MobileComm 近期熱門文章
PTT數位生活區 即時熱門文章
