|
|
of the product, while C-side product managers need to consider more about the ease of use of the product. The ability to gain life experience from the surrounding environment can be said to be the last private territory of human beings in the face of AI. But now, it seems that this private territory may be lost. Recently, DeepMind’s research results have discovered that a technology that has never existed has Agents that use any pre-collected human data can learn from scratch about the simulated environment around them and acquire human behavior. AI's ability to learn from the human world has so far remained at the language level.
Feeding a large model corpus—initially Wikipedia and Reddit, later ex Armenia WhatsApp Number panding to audio, visual images, and even radar and thermal images—the latter is, broadly speaking, another language of expression. Therefore, the entrepreneurs of generative AI believe that an extremely smart large language model is the final answer to AGI, and the multi-modal research path is currently insufficient for the former. Our imagination of unknown life forms is limited to this if silicon-based life counts. When talking about alien life, the first thought that comes to mind is alien language.
The first appearance of the Trisolaran in The Three-Body Problem is also about language. If we apply it to others, language will also be the operating system of other civilizations. Yuval Harari, the author of A Brief History of Humankind, publicly expressed his concerns about generative AI in May this year. AI that has mastered human language has the ability to hack into the entire human civilization. However, AI’s occupation of human language resources is also the limit of humanity’s current threat to AI. In other words, AI cannot learn things that cannot be expressed and recorded in language. The world is full of stories about scholars encountering soldiers. Reading thousands of books is not as good as traveling thousands of miles.
|
|
發表於 11:28:08