西西河

主题:【原创】深入谈一下chatGPT对我们我国的重大影响 -- 东东山

共:💬200 🌺1876 🌵20 新:
全看分页树展 · 主题 跟帖
家园 两个看法 -- 有补充

一,真正在复杂世界做决策和承担后果的工作职位不会受到AI的冲击,特别是那些必需人与人之间直接打交道的。AI对语义数据建构数学模型,对语义其实并没有人的理解(对抽象知识和逻辑投射的客观本体也不在乎,请注意,人的主观能动性也在于此)。物理事件的感知与初步分析必须依赖人,具体方案的系统规划与执行也必须依赖人。

二,哪怕AI掌握再多的自然语言数据,由于tokenization过程限制了模型的广泛覆域(计算量始终是瓶颈),对于长尾的知识领域必然缺位。此外看chatgpt全称,a generative pre-trained transformer , 如让它自己来阐释,pre-trained transformer 很好解释,什么是generative, 真正有新知的产生能力吗?

我的问:You are named a generative pre-trained transformer. Can you explain what is generative? Can you give some examples that AI can be generative as good as human being.

Chatgpt的答:

As a generative pre-trained transformer, I am capable of generating new content based on patterns learned from large amounts of text data. In other words, I can create new text that follows the same style and patterns as the data that I was trained on. "Generative" refers to the fact that I can produce new, previously unseen content that fits within the parameters of what I was trained on.

AI can be generative in a wide range of applications, including:

1. Text generation: As a language model, I can generate new text, such as product descriptions, news articles, or even poetry.

2. Music generation: AI can generate new music that is stylistically similar to a particular genre or artist. This can include melody, lyrics, and even instrumentals.

3. Art generation: AI can generate new art, such as images or even entire movies, based on patterns learned from existing data.

4. Game generation: AI can generate new levels, characters, and game mechanics for video games.

5. Language translation: AI can translate text from one language to another, generating new text in a different language that accurately conveys the meaning of the original text.

Overall, AI can be generative in many applications where pattern recognition and generation are important, and it can be as good as humans in generating content within specific parameters. However, AI still has limitations and is not yet able to fully match the creativity and originality of human beings in many domains.

我认知的chatgpt回答不过是以语言为对象的数据统计游戏。

维特根斯坦说:“命名似乎是一个词同一个对象之间的一种奇特的联结。——而且当哲学家盯着他面前的对象并且多次地重复一个名称或者甚至只重复‘这’这个词,企图以此来揭示名称和事物之间的这种关系时,你真的就得到了这种奇妙的联结。因为,当语言休假时,哲学问题就产生了。在这里,我们可能真的会幻想命名是心灵的某种奇异的活动,好像是对一个对象施行的洗礼。而且我们好像还能够对这个对象说‘这个’一词,用它去称呼它——对这个词的这样一种奇怪的使用,无疑地只有在搞哲学的时候才会发生。”

老子对这一主题的阐释是: 道可道,非常道。名可名,非常名。

关键词(Tags): #chatgpt#statistical game theory通宝推:方平,赵美成,陈王奋起,南寒,唐家山,
作者 对本帖的 补充(1)
家园 光说不练假把式 -- 补充帖

https://github.com/BlinkDL/ChatRWKV

这个开源项目可以git clone一下,试一下手。

另外,也可以试试以下:

chatgpt 与midjourney

影视业从业者AI指南chatgpt+midjourney

http://b23.tv/CBg5yht

midjourney AI绘图超实用指令

http://b23.tv/MqOBzdF

也可自行在youtube参考别人的使用来试

*** 规训的空间终究有限,与真实的复杂世界有别

庄子的逍遥游有句:

且夫水之积也不厚,则其负大舟也无力。覆杯水于坳堂之上,则芥为之舟;置杯焉则胶,水浅而舟大也。风之积也不厚,则其负大翼也无力。

通宝推:陈王奋起,
全看分页树展 · 主题 跟帖


有趣有益,互惠互利;开阔视野,博采众长。
虚拟的网络,真实的人。天南地北客,相逢皆朋友

Copyright © cchere 西西河