Here is the translation:
This is an article about the future trends of AI development. The article introduces the super intelligent possibility mentioned by Ilya in his speech, which includes:
1. The emergence of super intelligence will bring new abilities and characteristics with incredible capabilities.
2. Super intelligence can understand things from limited data without being puzzled.
3. Self-awareness is also useful, constituting a part of ourselves and our world model.
4. When all these features are combined with self-awareness, it will bring properties and characteristics completely different from existing systems.
The article also raises several questions:
1. Are there other biological structures belonging to human cognition worth exploring? Ilya thinks this depends on the abstract level we see.
2. Can model automatic correction in the future mean it can identify when illusions appear? Ilya gave a positive answer.
3. Can large language models generalize beyond distributed multi-hop reasoning? Ilya did not give a direct answer.
Finally, the article mentions that Google bigwig Rogen Killepatrick and Drewfus Batria, former Meta embodied intelligence team senior director, disagreed with Ilya's conclusions.
Translation
这是一篇关于未来AI发展趋势的文章。文章介绍了Ilya在演讲中提出的超级智能可能带来的变化和挑战,其中包括:
- 超级智能的出现将会带来新的能力和特性,它们将具有令人难以置信的惊人能力。
- 超级智能将能够从有限的数据中理解事物,并不会感到困惑。
- 自我意识也是有用的,它构成了我们自身的一部分,也是我们的世界模型中的一部分。
- 当所有这些特性与自我意识结合在一起时,将会带来与现有系统完全不同的性质和特性。
文章也提到了几个问题:
- 是否有其他属于人类认知的生物结构值得去探索?Ilya认为这取决于我们所看到的抽象层面。
- 模型的自动纠错在未来意味着它能够识别什么时候出现了幻觉?Ilya给出了肯定的答复。
- 大语言模型能否泛化超出分布的多跳推理?Ilya没有正面回答这个问题。
最后,文章提到了谷歌大佬洛根·基尔帕特里克和前Meta具身智能团队的高级总监德鲁夫·巴特拉对Ilya结论的不同意。
Reference:
https://www.youtube.com/watch?v=1yvBqasHLZs