AI泡沫应该会比dot.com泡沫爆得还狠吧

c
cloudy
回复 10楼 白河之子 的帖子
不是,统计正常生活中有用,但人类文明突破很多是靠outliers实现的,这正是被统计机制忽略不计的。
fantancyseeker 发表于 2025-01-05 14:18

所以可以用outlier的数据train高级AI 用普通人数据train普通AI?
未来高级AI负责科技创新 文明的迭代 普通AI 做人类的育儿嫂, 从出生到死亡(前提是人类还有存在的价值的话)
论AI阶级社会的诞生 LOL
c
cloudy
Microdisney 发表于 2025-01-05 14:56
ai是用来代替人类的,不是用来推动人类文明的。outlier这种话只有在进luigi楼喊口号的时候用,对ai不适用。

有可能
人类只是给更高级文明的出现搭个桥, 毕竟人这种生物要突破各种物理极限太难了
O
OroMedonte
刚看了眼这个帖子,手机谷歌就立马推送了这篇文章,AI 无处不在。
https://glassalmanac.com/the-google-expert-who-predicted-the-iphone-sees-a-future-where-everyone-lives-in-abundance-by-2030/
这个谷歌AI 总设计师连时间都推算出来了,2045年左右随着脑机接口技术的突破。太让人焦虑了。





f
fantancyseeker
回复 92楼 Iwkhx 的帖子
把人类智能简化为一堆参数基础上的统计推理,我认为是错误的。AI最多也就算是对人类智力的增强,本身不会做出什么突破性的发明创造。
g
gokgs
fantancyseeker 发表于 2025-01-05 14:29
回复 60楼 meftw 的帖子
没错,现在AI用DNN,跟以前的NN没本质区别,就是算力大了,可以更多层数和变量而已。

量变导致质变。 高中政治课白学了?
Hinton 的 文章 80 年代发布的。 为什么突然起飞了?
半个马和甲
回复 88楼 的帖子
果然AI和AI不一样。 这是我和他玩了10局后且全部完胜我写出的第一部分的逻辑分析

如此详细的分析建立自己的algorithm 背后,这个AI还有5段其他分析。 包括了: pattern recognition , next move recognition , counter move selection, adaptive strategy update and randomization for uncertainty.
这样的逻辑”思考”严密程度应该是完胜多数人类了。 这样的思考后, 我只要以一个rational human being 和它玩基本就越来越没有胜算了。算不过AI的 因为掌握这些以后,AI就”学会了”从第二局就开始布局,引导你(game theory)出对他有利的下一步。 他算得太快了,所以最终的结果就是你你为你是只猴子翻了十万8千里,而其实一直在它手里。 嗯。 (当然前提是我以rational thinking 和它玩。完全放弃大脑思考,开始混乱随机,它就懵圈了。 但人类不可能真正做到混乱的)
g
gokgs
OroMedonte 发表于 2025-01-05 15:16
刚看了眼这个帖子,手机谷歌就立马推送了这篇文章,AI 无处不在。
https://glassalmanac.com/the-google-expert-who-predicted-the-iphone-sees-a-future-where-everyone-lives-in-abundance-by-2030/
这个谷歌AI 总设计师连时间都推算出来了,2045年左右随着脑机接口技术的突破。太让人焦虑了。






too early to tell. but yeah, AI has huge potential.
智能机器人成功了, 就能利用智能机器人造机器人, 类似于核反应的 chain reaction, 人类文明指数级大发展。

c
cloudy
OroMedonte 发表于 2025-01-05 15:16
刚看了眼这个帖子,手机谷歌就立马推送了这篇文章,AI 无处不在。
https://glassalmanac.com/the-google-expert-who-predicted-the-iphone-sees-a-future-where-everyone-lives-in-abundance-by-2030/
这个谷歌AI 总设计师连时间都推算出来了,2045年左右随着脑机接口技术的突破。太让人焦虑了。






有什么好焦虑的, 我觉得挺兴奋的
现在就是AI的singularity, 能活着看到这个时代不容易啊
半个马和甲
Fritz2022 发表于 2025-01-05 13:56
回复 88楼 qgp 的帖子
我也试了和它玩了5局,我赢了4次,它赢1次。 我问它用的什么策略,它说:
其实我没有固定的策略,只是随机选择出石头、剪刀或布。你可以尝试根据我的出法进行预测或改变你的选择。你有没有特别的策略或想法?

你用的什么版本?! 我的AI 完爆我。写下来的策略,分析是真的非常全面了。
半个马和甲
fantancyseeker 发表于 2025-01-05 14:10
现在的AI,LLM是以对人类智能的误解为基础的。

I agree . 我也觉得AI要走远不是LLM 。 但这只是一个如第六感的东西。 可AI一定可以走远。因为我感到这是越来越靠近神造世界的奥秘。
g
gokgs
I agree . 我也觉得AI要走远不是LLM 。 但这只是一个如第六感的东西。 可AI一定可以走远。因为我感到这是越来越靠近神造世界的奥秘。
半个马和甲 发表于 2025-01-05 15:37

早着呢, 人类对灵魂的认知, 还基本是零, 连个起步都没有。
现在的人工智能, 还是物理器件堆积的结果, 即使成功了, 也还是个没有灵魂的工具, 跟阿猫阿狗还是有本质的区别。 阿猫阿狗都有爱恨情仇, 机器人不会有, 顶多有类似于 bug 的胡乱行动。
f
fantancyseeker
回复 103楼 gokgs 的帖子
政治课是真理?相对论是量变发现的?
g
gokgs
fantancyseeker 发表于 2025-01-05 16:01
回复 103楼 gokgs 的帖子
政治课是真理?相对论是量变发现的?

你的逻辑真是差到家了。
量变会导致质变, 不等于所有质变都是量变来的。 sigh, 你小学毕业了吗?
话说回来, 狭义相对论没有爱因斯坦, 应该就是晚了几年, 会有人发现, 广义相对论就只能靠老爱了。 当然, 这些对你来说都超纲了, 跟你说你也不会懂。
f
fantancyseeker
gokgs 发表于 2025-01-05 16:05
你的逻辑真是差到家了。
量变会导致质变, 不等于所有质变都是量变来的。 sigh, 你小学毕业了吗?
话说回来, 狭义相对论没有爱因斯坦, 应该就是晚了几年, 会有人发现, 广义相对论就只能靠老爱了。 当然, 这些对你来说都超纲了, 跟你说你也不会懂。

你这两头堵的话,不是放屁吗?哈哈哈哈😄,还小学毕业不毕业的,你啥毕业呀?让俺瞻仰一下。
g
gokgs
fantancyseeker 发表于 2025-01-05 16:17
你这两头堵的话,不是放屁吗?哈哈哈哈😄,还小学毕业不毕业的,你啥毕业呀?让俺瞻仰一下。

算了, 你智商太低了。 你自己牛吧。
平明寻白羽
我又把deepseek 给整疯了, lol
"write 15 words begin with letter a


**Spatial and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and **T cells, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, andFibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and) **Spatial and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and) **Spatial, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fib "
尽管write 15 words begin with letter a不是语法上完整的 “Write 15 words that begin with the letter "A"", AI 作为语言模型也不应该有问题的。
他发疯可能是我是在一个长chat里突然脱离context提了一个random task, 不知道怎么回答, 然后就进入了死循环。我不按stop, 他还会继续刷屏刷下去。 LOL

I
Iwkhx
fantancyseeker 发表于 2025-01-05 15:19
回复 92楼 Iwkhx 的帖子
把人类智能简化为一堆参数基础上的统计推理,我认为是错误的。AI最多也就算是对人类智力的增强,本身不会做出什么突破性的发明创造。

同意所有基于数学的算法都是加强性质的,效果是更快速,更准确。问题是人工智能在学习了足够的数据后,会不会自己产生出人意料的方法。出人意料是指不是人已经想到的方法让它去验证。比如用人工智能可以筛选药物化学成分。它有一天能不能告诉人类,这种筛选的路子就不对,应该是一种全新的人根本没有想到过的方法。
g
gopokemon
平明寻白羽 发表于 2025-01-05 16:25
我又把deepseek 给整疯了, lol
"write 15 words begin with letter a


**Spatial and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and **T cells, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, andFibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and) **Spatial and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and) **Spatial, and Fibfuture, and Fibfuture, and Fibfuture, and Fibfuture, and Fib "
尽管write 15 words begin with letter a不是语法上完整的 “Write 15 words that begin with the letter "A"", AI 作为语言模型也不应该有问题的。
他发疯可能是我是在一个长chat里突然脱离context提了一个random task, 不知道怎么回答, 然后就进入了死循环。我不按stop, 他还会继续刷屏刷下去。 LOL


如果和一个故意捣乱的人对话,估计大部分人都会失控吧。引入对这种非正常的行为的对应应该就好了吧?
f
fantancyseeker
Iwkhx 发表于 2025-01-05 16:34
同意所有基于数学的算法都是加强性质的,效果是更快速,更准确。问题是人工智能在学习了足够的数据后,会不会自己产生出人意料的方法。出人意料是指不是人已经想到的方法让它去验证。比如用人工智能可以筛选药物化学成分。它有一天能不能告诉人类,这种筛选的路子就不对,应该是一种全新的人根本没有想到过的方法。

就现在这种以DNN为底层的AI,我认为突破的可能性几乎没有。
d
diodiodio
回复 1楼 平明寻白羽 的帖子
ai不是泡沫 上限比dot com高太多了 并且未来三年内社会必然由于ai大变
M
Moscow79
回复 117楼 fantancyseeker 的帖子
问题是人脑本质上也就是一个训练了一亿多年的DNN,训练集特别大所以娃一生下来就有个很好的初值,用一点新数据fine tune一下就能用了
S
Soothsayer
回复 55楼 UncleSammy 的帖子
你说的这些现在都能做,但法规方面落地是个问题。 比如养老机器人,如果出了什么差错,谁负责。 公司不负责没人买,负责被 abuse 怎么办,问题一大堆。 要知道最早发明机器人帮自己养老的科学家就被自己的机器给绞死了。。。
很多产品,技术可行不意味着商业可行,跟政策一样,现实中的幺蛾子太多了
k
kengdie
AI 是泡沫,但是华尔街找到了bubble 不破的方法,就是不断创造相关的泡沫。大家使劲投钱造垃圾吧,啦啦啦啦啦啦
g
gokgs
回复 1楼 平明寻白羽 的帖子
ai不是泡沫 上限比dot com高太多了 并且未来三年内社会必然由于ai大变
diodiodio 发表于 2025-01-05 17:37

是的, 总的来说 AI 是人类的永远的追求。 chatGPT 出来也就两年吧, 也许至少再住 5 年, 才能对这轮发展有个定论。
中间也许会有人掉队, 大方向应该不会改变。
Conclusion: AI might be overhyped in some areas, especially when it comes to expectations about its capabilities in the short term (e.g., "AGI will come in the next 5 years!"), but calling it a "bubble" may oversimplify the situation. There’s undeniable, growing value in AI technologies that are being developed and deployed right now. Whether it’s a "bubble" will depend on how well companies can manage expectations, deliver on their promises, and continue to innovate in meaningful ways. If the hype outpaces the real-world impact, then yes, it could experience a "bubble-like" correction—but the underlying technology will likely continue to thrive in the long run. So, it might be more accurate to say that AI is both a revolution and a speculative market, with some aspects of it carrying bubble-like risks, but with foundational technologies that are likely here to stay and evolve.
c
cloudy

系统提示:若遇到视频无法播放请点击下方链接
https://www.youtube.com/embed/s6T4DXRKYHM
f
fantancyseeker
Moscow79 发表于 2025-01-05 17:45
回复 117楼 fantancyseeker 的帖子
问题是人脑本质上也就是一个训练了一亿多年的DNN,训练集特别大所以娃一生下来就有个很好的初值,用一点新数据fine tune一下就能用了

所以你认为人脑的工作机理已经完全搞清楚了?就是一个简单的DNN?