Articles

Concept

Thinking = Model = Knowledge

Definition of knowledge: An expression of information that can be used to solve multiple tasks.

 

Generalization: The ability to infer from one instance to another is generalization.

Language brings us the fundamental ability, which is zero-shot generalization. Communication is just a by-product of language.

Humans have language, hence they have a second system and think. 

Paradigm Shift

Tripartite evolution mode: Perception (information system) → Thinking (model system) → Implementation (action system)

  

The shift in the new paradigm is fundamentally a shift from marginal cost to fixed cost.

In 1995, there was a paradigm shift in information retrieval. Google only needed to pay a fixed $1 billion to provide information retrieval services to the whole world.

In 2023, there was a paradigm shift in knowledge acquisition. Large models can think now; model equals knowledge, and knowledge and cognitive intelligence will be ubiquitous (represented by OpenAI). We are currently in the Cambrian era where large models give rise to sub-models.

The next phase should be the action system, with robots everywhere (represented by Tesla).

Opportunity

Economic paradigm shift: Agricultural economy → Commodity economy → Service economy → Experience economy → AI economy

Shift in representative occupations: Farmer → Worker → Coder, designer, analyst → Entrepreneur, scientist → New human value system

In the next era, the representative occupations will shift to entrepreneurs and scientists. The core ability will no longer be knowledge, but insight and creation.

 

Opportunities for entrepreneurs:

"Information-Model-Action" tripartite is the endgame. If there's an opportunity for the tripartite, one must seize it first; have your own data; can develop your own model; data-model-use-data closed loop: after the model is used, data can flow back, empowering proprietary models.

 

In the era of prioritizing large models, how to choose innovative opportunities at the application level? Technologically:

1. In your field, is the progress of the model fast, medium, or slow? Prioritize the fast ones.

2. Opportunities for the tripartite experience brought about by the large model, is it high, medium, or low? High, for example, advertising and marketing.

3. Has the large model brought about breakthrough progress in the R&D system of the field?

 

Besides, each field has a focus on the tripartite point. For example:

Finance focuses on information.

Search and advertising focus on the model. Because over half of the search demand is for knowledge acquisition rather than simple information retrieval, it's about acquiring a model. Computational advertising is about finding the right person in the right scenario. Informationalization was completed in the first phase, and it's now in the upgrade from perceptual model to cognitive model.

Fitness focuses on action.

 

Games: Large models have a significant disruption to games because every link from content production to experience in games can be reconstructed and enhanced by large models. Games are the digital twins of the world.

 

Social networks: If you have good content (model), you have the opportunity to establish a new social network. — For example, TikTok.

My Insights

The advertising and marketing system will be overturned

The existing product and service recommendation system: products and services find people mainly, people match with products. Rely on collaborative filtering, similar recommendations, or manually summarizing and tagging to recommend products to people.

People not only have to deal with various information interferences but also are responsible for various purchasing comparison decisions. The richer and more varied the supply of goods and services, the higher the cost of this purchasing decision.

In the future, the paradigm should shift to large models thinking mainly, where people only need to trade background information and make key decisions.

Humans will leave daily energy-consuming thinking, non-professional decisions to large models, and everyone will have their own personal secretary. There's no need to shop on e-commerce platforms, and no need for advertisements. Merchants' credit and product usage feedback will be automatically written by large models and uploaded to the public chain.

In fields that require deep thinking and heavy decision-making, such as investment and home decoration, the disruption will be smaller.

Branding and positioning theories will also be weakened

The core assumption of the positioning theory is that consumers only remember one keyword in each niche. But if large models take over purchase decisions, this assumption no longer exists.

Brands exist to save the comparison selection cost in transactions and reduce the risk of breach of trust. But what if large models make comparison selection costs extremely low, and blockchain makes the risk of breach of trust extremely low?

The GDP share of digital content further increases

When information-knowledge becomes a fixed cost, the bottleneck will shift to material (resources) and energy.

As more and more people are freed from physical labor and medium-to-low-level mental labor, humans to some extent can meet the basic needs of survival without working, and the remaining pursuit is experience (in plain terms, the secretion of dopamine higher than the baseline level). While the experience of physical objects is limited by material resources, games and digital content, as twins of the world, offer richer and cheaper experience possibilities and will become the mainstay of the experience economy.

How to choose a field to enter?

Top-tier big cake: consumer market, online advertising, precise recommendation, e-commerce.

Long-tail market: 2B market, various niche needs.

The more essential the need, the more urgent the need for the tripartite. For example, catering, medical care.

Also, consider the progress of DIKW in this field.