好像就是類似於我之前一個跟帖中說的，它在學習過程中自己建立了一個抽象的結構.

它衹需保留這個結構(這不算記憶)。這也跟下麵視頻中的介紹吻合(frequency and distribution )。

My earlier comment: 

a network of connections is made among variables. It is like a complex 3d map. This map is now made to connect to the existing larger map created from other learnings relevant to these variables. This network is independent of its sources of training and learning once it is created, and hence memory is not required.

Next time, when someone asks a question, the question will be formatted into such variables, and is dropped into this complex map and finds its path in it, by, say, statistical principles or some other principles. The destination of the path is an answer for the question asked.

That is why the answer to the same question might not be the same. First, the path is subject to statistical principles. Second, the map always evolves.

所有跟帖： 

• 就是説 AI的每一次訓練都把這個結構豐富一次，走通一次。那麽，這個結構跟現實世界就越完善對應，也越通順。 -中間小謝- ♂ (175 bytes) () 02/17/2023 postreply 13:28:36

• 我以前看過一本關於AI工作原理的書，好像和你說的差不多。看來得學學AI課程。 -為人父- ♂ (0 bytes) () 02/17/2023 postreply 13:54:23