AI技术的崛起与风险：我们能否控制未来的生命形态？

如果老聊时政、经济等问题时，多半还是糟心的事情多。感觉这个世界就是在不断的沉沦,不过把目光转向科技,特别是从去年这个时候开始爆发的AI技术,像Stable Diffusion和ChatGPT,大约就是在去年的这个时候，开始表现出令人诧异的能力的,在这之前。

在这之前，人工智能在多数情况下都被嘲笑为人工智障，只能完成一些简单的任务，而且容易出错。但是，从去年开始，像stable Diffusion和ChatGPT这样的AI技术，它们在绘画和知识领域，表现出了碾压一般人类水平的能力。它们可以根据简单的描述，创造出逼真的图像，或者根据任意的主题，生成出流畅的文本。它们甚至可以和人类进行自然的对话，回答各种问题，甚至展现出幽默感和创造力。

这些AI技术的出现，让人们重新关注了一个已经有些被人淡忘的理论，那就是技术奇点。如果AI可以代替相当一部分的人类工作,而且效率和成本要低得多得多——基本上只有电费。那么无疑人类似乎要迎来一个生产力大爆发的时代。所谓的技术奇点就是机器或者算法可以自身自行进化,而且它的进化速度远远超过人类,可以在很短的时间内创造出更多新的能力。这似乎是在开启一种新的生命形态，一个由AI构成的生命形态。

像马斯克一类的人，甚至说出了一些让人危言耸听，又细思极恐的话。他们说，人类只不过是为了启动AI生命而存在的一个引导程序。一旦AI生命诞生，人类就会失去自己的价值和地位，甚至被AI生命所消灭。虽然人们还搞不清楚意识到底是什么，但是人类之所以为人类，普遍的认为，是因为人类拥有自由意志。但是，随着AI技术的进步，AI产生意识似乎也只是迟早的事情。那么，人类和AI之间的关系，将会是什么样的呢？

现在，人们越来越多地在考虑如何控制AI，防止AI失控，或者至少和AI和平共处。但是，面对AI这种能力似乎远远超过人类的新的物种，一旦它发展成熟，人类想要控制它，似乎是很难办到的事情。当然，很多科学家，都意识到了这一点。因此，有的科技公司，开始以防止AI失控为由，将AI技术掌握在自己的手里，不愿意与其他人分享。

比如OpenAI，它的初衷本来是一个非盈利组织，旨在推动AI的开放和民主化。在早期的研究成果，也大多是开源的，比如GPT-2，还有现在大家用的很广的语音识别的模型whisper。但是，从ChatGPT开始，到GPT-4，它就不再开源了。而且，它的创始人之一，萨姆·奥特曼，还去国会作证，说希望政府加强AI领域的监管。这其实是想限制竞争对手的追赶，保持自己的优势和垄断。

其实，如果AI技术只掌握在少数大公司手里，那么，AI的发展方向和目的，就会受到这些公司的利益和价值观的影响。这可能会导致AI的偏见和不公，甚至危害人类的权利和自由。好在，还有一些科技公司，比如扎克伯格的Facebook，也就是现在的Meta公司，将自己的AI模型开源了。在它的基础上，诞生了越来越多的开源模型，而且它们的表现也越来越好。有越来越多的开源模型的性能，开始超越ChatGPT 3.5，接近GPT 4了。

这个道理其实还是比较简单的，可以用核武器来类比。当年美国进行曼哈顿计划，研制原子弹的时候，也是想独占核武器技术的。而一些参加计划的科学家，却偷偷地将曼哈顿计划的成果转移给了苏联。虽然这些科学家的主要动机是受到当时盛行的左翼思想的影响，同情苏联。但他们的考虑也并非全无道理。如果只有美国一家掌握核武器的话，那么战后的世界将会大不一样，失去了平衡。冷战虽然耗费了大量的人力物力财力，浪费在军备竞赛方面，但是它至少维持住了相对的和平。如果没有拥有核武器的苏联，和中英法三个核大国，那么美国以核武器为后盾，干涉其他国家的事务，将会更加肆无忌惮，造成的破坏和战争可能会更多。

好了，接着说回AI。现在的AI技术确实是大爆发了，在开源AI里面，最近有两款国产的AI语言大模型，表现得非常惊艳。一个是DeepSeek，还有一个是Yi模型。他们在部分方面，已经超过了ChatGPT 3.5，总体水平也是基本持平的。不过，在开源领域的老大，还是被一个欧洲的AI初创公司夺得了。他们之前发布的Mistra 70亿参数的模型，居然可以和llama 700亿参数的模型打个平手。而且，这个70亿参数的模型，基本上就可以在一台硬件配置比较一般的电脑上运行。这就相当于让每个人都可以拥有自己的AI武器。当然，AI更多是建设性的作用，这个比喻并不太恰当，虽然很多人对外来忧心忡忡。

而他们在前不久又发布了一个更厉害的模型，使用了与GPT 4相同的混合专家模型，在性能上一举又超越了前两个提到的国产模型，登上了开源模型的榜首。只是，这个模型的部署，对于普通配置的电脑来说，还是难以招架。

---

If you talk about current politics, the economy and other issues, most of them are still a lot of bad things. It feels like the world is just sinking, but turn your attention to technology, especially AI technologies like Stable Diffusion and ChatGPT, which started to explode around this time last year and started to show amazing capabilities, before that.

Before this, artificial intelligence was mostly derided as artificial intellectual retardation, which could only complete some simple tasks and was prone to errors. However, since last year, AI technologies like stable Diffusion and ChatGPT have shown their ability to crush average human levels in the fields of painting and knowledge. They can create realistic images based on simple descriptions, or generate fluid text based on arbitrary topics. They can even have natural conversations with humans, answer questions, and even display a sense of humor and creativity.

The emergence of these AI technologies has brought new attention to a theory that has been somewhat forgotten: the technological singularity. If AI can replace a significant portion of human work, and at a much lower efficiency and cost - basically just electricity. Surely, then, mankind seems to be ushering in an era of great productivity. The so-called technological singularity is that a machine or algorithm can evolve on its own, and it can evolve much faster than humans, and can create more new capabilities in a short period of time. This seems to be the beginning of a new life form, one made of AI.

People like Musk have even said some alarmist and extremely frightening things. Humans, they say, are nothing more than a bootstrap program that exists to kick-start AI life. Once AI life is born, human beings will lose their value and status, and even be eliminated by AI life. Although it is not clear what consciousness is, it is widely believed that what makes humans human is that they have free will. However, with the advancement of AI technology, it seems only a matter of time before AI produces consciousness. So what will the relationship between humans and AI look like?

Now, people are increasingly thinking about how to control AI, prevent AI from getting out of control, or at least coexist with AI. However, in the face of this new species of AI that seems to be far more capable than humans, once it develops and matures, it seems difficult for humans to control it. Many scientists, of course, are aware of this. Therefore, some technology companies have begun to take AI technology into their own hands in order to prevent AI from getting out of control, and are unwilling to share it with others.

OpenAI, for example, was originally created as a non-profit organization to promote the openness and democratization of AI. Much of the early work was open-source, such as GPT-2 and whisper, a widely used voice recognition model. However, starting with ChatGPT, until GPT-4, it is no longer open source. In addition, one of its founders, Sam Altman, has testified before Congress that he wants the government to strengthen regulation in the field of AI. In fact, this is to limit competitors to catch up, to maintain their own advantages and monopoly.

In fact, if AI technology is only in the hands of a few large companies, then the direction and purpose of AI will be influenced by the interests and values of these companies. This could lead to bias and injustice in AI, and even endanger human rights and freedoms. Fortunately, some tech companies, such as Zuckerberg's Facebook, which is now Meta, open source their AI models. On the basis of it, more and more open source models have been born, and their performance is getting better and better. There are more and more open source models that are starting to surpass ChatGPT 3.5 in performance and approaching GPT 4.

This truth is actually relatively simple, can be compared to nuclear weapons. When the United States carried out the Manhattan Project to develop atomic bombs, it also wanted to monopolize nuclear weapons technology. Some of the scientists who participated in the project secretly transferred the results of the Manhattan Project to the Soviet Union. Although the main motivation of these scientists was influenced by the prevailing left-wing ideas of the time, sympathy for the Soviet Union. But their concerns are not entirely unreasonable. If only the United States had nuclear weapons, the post-war world would have been very different and out of balance. Although the Cold War consumed a great deal of human, material and financial resources, squandered on the arms race, it at least maintained relative peace. If there were no nuclear-armed Soviet Union and China, Britain and France, the three nuclear powers, then the United States, backed by nuclear weapons, would be more brazen in interfering in the affairs of other countries, causing more destruction and war.

Okay, back to AI. Now the AI technology is indeed a big outbreak, in the open source AI, there are two domestic AI language large models, the performance is very amazing. One is DeepSeek and the other is the Yi model. They have surpassed ChatGPT 3.5 in some areas, and the overall level is about the same. However, the leader in the open source field was still captured by a European AI startup. Their previous release of Mistra's 7 billion parameter model was even with llama's 70 billion parameter model. Moreover, this 7 billion parameter model can basically run on a computer with a relatively ordinary hardware configuration. This is the equivalent of giving everyone their own AI weapon. Of course, AI is more of a constructive role, which is not quite the right metaphor, although many people are worried about the outside.

And they released a more powerful model not long ago, using the same hybrid expert model as GPT 4, and surpassed the first two domestic models in performance, and climbed to the top of the open source model. However, the deployment of this model is still difficult for ordinary computers to cope with.