Try to compare humans and AI to improve understanding of each other.
The foundation of this wave of large models - Transformer, has a great feature, that is, "serial to parallel", which is more like the operating mode of the human brain.
The source of AI capabilities, from the bottom-level large model to the top-level prompt, has many layers, for example, a middle layer is fine-tune. In analogy to the source of human abilities, the bottom layer is the innate brain, and then the acquired environment. Small, medium, and university masters and Ph. .
Therefore, every time a large model is trained, it is a bit like creating an AI baby? In addition, the direction of education in the later period will be different, and AI will flourish in the future, including liberal arts AI, science AI, AI for various positions, and so on.
Illustrations for AI imitating paintings drawn by my son
The emergence of human wisdom is thought first, then language (realizing inter-individual interaction, spanning space), and finally writing (realizing intergenerational communication, spanning time). Thought-language-text, and the birth of AI wisdom, the text is presented as soon as it comes up. After the text, the language is almost simultaneously produced, and the thought, everyone thinks, will be born last. Is the generation of AI wisdom reversed, or is human understanding wrong?
Next, for human beings, they can think more than they can speak, and they can speak more than they can write. Now, AI is trained by the written information. In the future, it is conceivable that the written information is directly intercepted from the source through the input method, or WeChat directly produces a small model, and directly accesses my dialogue + official account + circle of friends, etc.; the stage of speaking is portable radio Devices, things like Siri, and brain-computer interfaces that directly interact with human thoughts...seem to be an upgrade path.
The chain of thought is really confusing. It is wrong for you to ask AI to give the answer directly, but let it think step by step, and the result will be right. Analogy to humans, it seems that from "fast thinking" to "slow thinking", of course, additional computing resources are also required to be allocated to more reasoning steps.
Note: Refer to "Thinking: Fast and Slow", animal intuition vs human rationality.
Learning by example (few shot learning/prompt) is also very similar to human beings. In daily communication, if we don’t understand what the other party is saying, we often say-can you give me an example or an analogy?
The above analogy is actually a very typical thinking mode, that is, cross-border learning and knowledge transfer. It is difficult for human beings to be familiar with multiple fields at the same time, so they often use the fields they are familiar with to transfer and understand other fields. AI can be familiar with multiple fields at the same time, will it do better in cross-border inspiration?
Let's try it, such as user operations and human resource management. I have a little understanding of these two fields. There are many methods, frameworks, and theories for interacting with specific groups. There should be many points that can inspire each other.
The screenshot below shows the AI answer.
_________
Su Jie (iamsujie), product innovation consultant, author of 4 books in the "Everyone is a Product Manager" series, former Ali product manager for 8 years, head of the Group's product university, and founding partner of Liangcang Incubator.
If you need training and consulting services in product manager/product thinking/product innovation related fields, please contact this WeChat (13758212411).