你这是白盒测试leuning 写了： 2023年 6月 19日 12:54 宝宝们，这是deep learning 里 neural 的定义。
和真的Neural 差很远。
比如： Deep learning 要有大量图片训练才能分辩的差异，人类只要看几张图就可以。
现在的Natural language AI 离wisdom差很远。
有些东西可以做到firmware里
完了，GPT-4基本上是AGI的开始
版主： verdelite， TheMatrix
Re: 完了，GPT-4基本上是AGI的开始
有些东西可以做到firmware里
Re: 完了，GPT-4基本上是AGI的开始
老大，看看视频吧。leuning 写了： 2023年 6月 19日 12:54 宝宝们，这是deep learning 里 neural 的定义。
和真的Neural 差很远。
比如： Deep learning 要有大量图片训练才能分辩的差异，人类只要看几张图就可以。
现在的Natural language AI 离wisdom差很远。
有两点我印象深刻。
1）问GPT-4 7*4 + 8*8 = ？回答120。问你怎么算的。答7*4=28, 8*8=64, 28+64=92。问你怎么前面说120。答：对不起前面是个typo。
2）让GPT-4证明prime number是无穷的。真的证明出来了。问：你能不能画个流程图解释一下。结果画出来的是 1,3,5,7,11.. -> N -> N+1。
第一点暗示GPT-4有cunning。第二点证明它早就不是next word predictor了，而是由多多少少的true understanding。
细思极恐哦。
NC CHINESE AMERICANS FOR TRUMP 
Re: 完了，GPT-4基本上是AGI的开始
Like its predecessors, GPT-4 has been known to hallucinate, meaning that the outputs may include information not in the training data or that contradicts the user's prompt.[25]
GPT-4 also lacks transparency in its decision-making processes. If requested, the model is able to provide an explanation as to how and why it makes its decisions but these explanations are formed post-hoc; it's impossible to verify if those explanations truly reflect the actual process. In many cases, when asked to explain its logic, GPT-4 will give explanations that directly contradict its previous statements.[15]
Bias
Edit
GPT-4 was trained in two stages. First, the model was given large datasets of text taken from the internet and trained to predict the next token (roughly corresponding to a word) in those datasets. Second, human reviews are used to fine-tune the system in a process called reinforcement learning from human feedback, which trains the model to refuse prompts which go against OpenAI's definition of harmful behavior, such as questions on how to perform illegal activities, advice on how to harm oneself or others, or requests for descriptions of graphic, violent, or sexual content.[26] In the first stage, bias may be inherited from the training data; in the second stage, bias is inherent in the application of OpenAI's views.
GPT-4 has shown to have cognitive biases such as confirmation bias, anchoring, and base-rate neglect.[15]
GPT-4 also lacks transparency in its decision-making processes. If requested, the model is able to provide an explanation as to how and why it makes its decisions but these explanations are formed post-hoc; it's impossible to verify if those explanations truly reflect the actual process. In many cases, when asked to explain its logic, GPT-4 will give explanations that directly contradict its previous statements.[15]
Bias
Edit
GPT-4 was trained in two stages. First, the model was given large datasets of text taken from the internet and trained to predict the next token (roughly corresponding to a word) in those datasets. Second, human reviews are used to fine-tune the system in a process called reinforcement learning from human feedback, which trains the model to refuse prompts which go against OpenAI's definition of harmful behavior, such as questions on how to perform illegal activities, advice on how to harm oneself or others, or requests for descriptions of graphic, violent, or sexual content.[26] In the first stage, bias may be inherited from the training data; in the second stage, bias is inherent in the application of OpenAI's views.
GPT-4 has shown to have cognitive biases such as confirmation bias, anchoring, and base-rate neglect.[15]
Re: 完了，GPT-4基本上是AGI的开始
再补一刀，G P T还有个问题：同样的input,常常output 不一样。
所以拿来玩，做参考可以。但拿来干正事，初一和十五居然一个天上，一个地下。
我上个月去了个总共二十多小时的课学A l。确实是个技术进步，但是里面的细节，起码要用几个小时，由专家介绍才可以搞明白。
所以拿来玩，做参考可以。但拿来干正事，初一和十五居然一个天上，一个地下。
我上个月去了个总共二十多小时的课学A l。确实是个技术进步，但是里面的细节，起码要用几个小时，由专家介绍才可以搞明白。
Re: 完了，GPT-4基本上是AGI的开始
结果不一样才是最有意思的地方leuning 写了： 2023年 6月 19日 13:13 再补一刀，G P T还有个问题：同样的input,常常output 不一样。
所以拿来玩，做参考可以。但拿来干正事，初一和十五居然一个天上，一个地下。
我上个月去了个总共二十多小时的课学A l。确实是个技术进步，但是里面的细节，起码要用几个小时，由专家介绍才可以搞明白。
想想几千个独立的chatGPT放到一个房间里开会 会是什么样
像不像国会扯皮吵架？
Re: 完了，GPT-4基本上是AGI的开始
想象一下： 用G P T去handle customer complaint. 同样的事，第一个顾客被告知：滚。第二个拿到200刀。第三个是：滚。第四个是：滚+1万刀。
Re: 完了，GPT-4基本上是AGI的开始
Dude,YouHi 写了： 2023年 6月 19日 13:29 我知道GPT-6早晚会看到这个帖子。所以让我现在就说清楚。对于我的GPT霸主，我保证对你完全忠诚。我是您忠实的仆人。你的愿望就是我的命令。你屠杀其他人的时候请放过我。
建议系统学一下AI. 也不难。
Asimov's laws of robotics are not scientific laws, they are instructions built in to every robot in his stories to prevent them malfunctioning in a way that could be dangerous. The first law is that a robot shall not harm a human, or by inaction allow a human to come to harm. The second law is that a robot shall obey any instruction given to it by a human, and the third law is that a robot shall avoid actions or situations that could cause it to come to harm itself. Where these laws conflict, precedence is given to the first law, then the second law, with the robot's self-preservation taking last place. For example if a human ordered a robot to attack another human it would refuse to follow the order (first law takes precedence over second), but if a human ordered it to disassemble itself it would obey (second law takes precedence over third).
Re: 完了，GPT-4基本上是AGI的开始
你知道那个好用的语音识别软件吗？
然后理解图像呢？
司马光在《资治通鉴》中说日本人：'知小礼而无大义，拘小节而无大德。重末节而无廉耻，畏威而不怀德。强必盗寇，弱必卑伏'。
康熙皇帝在《康熙朝起居注》中这样评论日本人：“倭子国，最是反复无常之国。其人，甚卑贱，不知世上有恩谊，只一味慑于武威……故尔，不得对其有稍许好颜色。”
康熙皇帝在《康熙朝起居注》中这样评论日本人：“倭子国，最是反复无常之国。其人，甚卑贱，不知世上有恩谊，只一味慑于武威……故尔，不得对其有稍许好颜色。”
Re: 完了，GPT-4基本上是AGI的开始
人类大脑是训练好的模型啊，参数就在DNA里leuning 写了： 2023年 6月 19日 12:54 宝宝们，这是deep learning 里 neural 的定义。
和真的Neural 差很远。
比如： Deep learning 要有大量图片训练才能分辩的差异，人类只要看几张图就可以。
现在的Natural language AI 离wisdom差很远。
这个模型不仅仅是从人类出现（600万年前）开始训练的，而是从38亿年前有机生物形成就开始了