The Network Effects Manual: 13 Different Network Effects (and counting)

The Network Effects Manual: 13 Different Network Effects (and counting)
Perspectives in AI: From LLMs to Reasoning with Edward Hu, Inventor of LoRA and μTransfer

In the digital world, network effects (nfx) have become the number one way to create defensibility for tech companies. These effects, which are responsible for 70% of the value created by tech companies since the advent of the Internet, are crucial for companies to win big. Among the four remaining defensibilities in the digital age (brand, embedding, scale, and network effects), network effects are by far the strongest.

It's important to note that network effects are not the same as viral effects. While viral effects focus on getting new users for free, network effects are about creating defensibility. The value of a network increases exponentially in proportion to the number of users, as Reed's Law suggests. This exponential growth makes it difficult for new entrants to match the value provided by an established network.

Network effects can be seen in various types of networks, such as personal utility networks and marketplace networks. Personal utility networks provide practical utility to users and are often used for private communication. On the other hand, marketplace networks aggregate competing sellers in one location, providing more business opportunities than scattered sellers. Breaking apart these networks requires offering a better value proposition for both parties simultaneously.

However, both marketplace networks and platform networks are vulnerable to multi-tenanting. This occurs when users or app developers create versions of their products for multiple platforms, reducing the exclusivity and defensibility of the network. To combat this, platforms must design their product or service to add significant value or "lock-in" for members, preventing them from being tempted to multi-tenant.

Data network effects occur when a product's value increases with more data, and the additional usage of the product yields useful new data. The relationship between product usage and the amount of new data gathered can be asymmetrical. Technological advantages, although initially advantageous, have a short half-life and are not highly defensible in the long term. However, tech performance network effects can provide a runaway advantage for being the first to enter the market with a new technology.

Social network effects are challenging to deploy for long-term defensibility but can offer a significant advantage if successfully utilized. Belief network effects, also known as bandwagon effects, occur when beliefs become more valuable to believers as more people adopt them. Apple has successfully leveraged bandwagon effects with their carefully scripted product demos and launches, creating buzz and FOMO (fear of missing out) among consumers.

In the field of AI, Low Rank Adaptation (LoRA) is an effective method for adapting large, pre-trained models to specific tasks or domains without significant retraining. LoRA acts as an auxiliary component that adjusts the model's characteristics without requiring extensive rebuilding or retraining. By leveraging the concept of low rank approximation, LoRA creates a smaller, adaptable module that can be integrated into larger models to customize them for specific tasks.

The use of LoRA in AI models has led to impressive efficiencies. It allows for faster and smaller adaptations, reducing resource usage and storage costs. For example, the reduction in checkpoint sizes from 1 TB to just 200 megabytes enables innovative engineering approaches such as caching in VRAM or RAM. These advancements improve user experience and reduce training costs by decreasing the number of GPUs required.

In conclusion, network effects and LoRA have emerged as powerful tools in the digital world. Understanding and harnessing the various types of network effects can provide companies with a significant competitive advantage. Additionally, incorporating LoRA into AI models allows for efficient adaptation and customization, resulting in cost savings and improved performance. To leverage these concepts effectively, companies should focus on delivering value, creating exclusivity, and understanding the psychology of their target audience.

Actionable advice:

1. Identify the network effects relevant to your business and focus on building and strengthening them.
2. Explore the potential of LoRA or similar methods to adapt and customize AI models for specific tasks or domains, reducing resource usage and improving performance.
3. Understand the psychology of your target audience and leverage social or belief network effects to create a sense of exclusivity and drive adoption.

By incorporating these strategies, companies can enhance their defensibility, optimize resource usage, and create a compelling user experience.