Predicting machine learning moats is a complex task that requires tracking the interface between scaling laws and products. While software scales with zero marginal costs, machine learning scales with nonlinear emergent behaviors. In order for a business to have a truly great and enduring "moat," it must protect excellent returns on invested capital. The challenge lies in the fact that models can easily be replaced with fine-tuning or procedural changes. The model is the part of the system that users interact with the most, but it is the dataset, infrastructure, and processes that create structural advantages. Currently, data is the moat for ML systems. When training data is well-defined and curated over time, it becomes difficult for it to be taken by an employee leaving the company or through a simple leak. User data is especially valuable as it provides diverse and non-repeated data, which is crucial for scaling. By adding new data that leads to new abilities and concentrated usage, companies can create lasting advantages that were not seen before. Some companies, like Runway and Jasper, are successfully crafting moats in specific verticals where they have established themselves as the best-in-class companies and brand names. On the other hand, Lensa, which is built on Stable Diffusion, may not have a moat at all and may have only gained success by being the first in the market.

Predicting machine learning moats is a complex task that requires tracking the interface between scaling laws and products. While software scales with zero marginal costs, machine learning scales with nonlinear emergent behaviors. In order for a business to have a truly great and enduring "moat," it must protect excellent returns on invested capital. The challenge lies in the fact that models can easily be replaced with fine-tuning or procedural changes. The model is the part of the system that users interact with the most, but it is the dataset, infrastructure, and processes that create structural advantages. Currently, data is the moat for ML systems. When training data is well-defined and curated over time, it becomes difficult for it to be taken by an employee leaving the company or through a simple leak. User data is especially valuable as it provides diverse and non-repeated data, which is crucial for scaling. By adding new data that leads to new abilities and concentrated usage, companies can create lasting advantages that were not seen before. Some companies, like Runway and Jasper, are successfully crafting moats in specific verticals where they have established themselves as the best-in-class companies and brand names. On the other hand, Lensa, which is built on Stable Diffusion, may not have a moat at all and may have only gained success by being the first in the market.

The phenomenon of notes apps is interesting because it is where ideas go to die, and surprisingly, that is a good thing. We often save notes, links, ideas, and thoughts as a form of insurance for the future. By saving them, we feel safe, even though we rarely look back at all the notes later. It seems that we are constantly seeking the next best thing and the newest tools to organize our thoughts and ideas. However, the true value of these notes apps is not in remembering everything we write down but in forgetting. These apps serve as insurance for our ideas, allowing us to let go and move on. Most of our thoughts and random discoveries are not actually valuable in the long run. We may write them down, but we rarely give them a second thought. In fact, we could burn the paper and scatter the ashes, and it would provide the same value. The problem lies in our tendency to ascribe value to our thoughts and findings. We fear losing them because we want what we do to have meaning and to feel productive. This fear is rooted in loss aversion, a concept explained by Daniel Kahneman in his book "Thinking, Fast and Slow." Loss aversion is the phenomenon where the response to losses is stronger than the response to gains. It is a biological trait that has been naturally selected into our DNA. However, trying to remember everything and holding onto every thought and idea can lead to a mental burden and hinder our ability to remember the truly important things. We need to forget, but we first need to feel safe forgetting. We need to believe that our memories were not in vain and that they will be there if we need them again. Only then can we let go and make room for new ideas and discoveries.

The constant search for the next best notes app or tool is a reflection of our desire to find something better, something that will finally help us organize our thoughts and ideas. We blame the tools and techniques, thinking that a new app will be the solution. However, this cycle of seeking the new and better thing ultimately leads us back to where we started – safely forgetting things. It is a never-ending illusion of value that keeps us trapped in a cycle of constantly seeking and never truly finding. We keep adding our newest thoughts to the latest app, exploring its features and organizational capabilities, only to eventually realize that it is just another tool that cannot solve the fundamental issue of forgetting. We find ourselves disillusioned and on to the next new thing, hoping that it will finally be the solution.

In reality, the best ideas we come across are the ones that resurface time and time again. Our notes end up being a record of when we first encountered these ideas, rather than a means of truly capturing their value. However, in a world where storage is cheap and unlimited, we continue to hold on to this illusion of value. It gives us a sense of mental safety, knowing that we have recorded our thoughts and ideas, even if we rarely revisit them. We must recognize that we don't need to remember everything and that it is okay to let go of ideas that are not truly valuable. By embracing the act of forgetting and focusing on the important things, we can free ourselves from the burden of constantly seeking the next best thing in notes apps and tools.

In conclusion, predicting machine learning moats requires understanding the interface between scaling laws and products. Data is currently the moat for ML systems, providing lasting advantages when well-defined and curated over time. On the other hand, notes apps serve as insurance for ideas, allowing us to forget and move on. We often seek the next best thing in notes apps, hoping to find a solution to organizing our thoughts and ideas. However, the true value lies in embracing the act of forgetting and focusing on the important things. Here are three actionable pieces of advice to consider:

1. Prioritize data curation and infrastructure: To create a lasting advantage in the field of machine learning, focus on building a strong foundation of well-defined and curated data. This will provide structural advantages that cannot easily be replaced.

2. Embrace the act of forgetting: Instead of constantly seeking the next best notes app or tool, recognize that not all ideas and thoughts are valuable. Allow yourself to let go of ideas that do not serve a purpose and focus on the important ones.

3. Seek meaning and value in the important things: Rather than ascribing value to every thought and finding, focus on finding meaning in the ideas and discoveries that truly matter. Let go of the illusion of value and embrace the idea that not everything needs to be remembered.

By incorporating these actionable pieces of advice into your approach to machine learning moats and notes apps, you can navigate the complexities of both fields and find value in the most important aspects.