AI intelectual property protection. But actually it’s about learning!
Is this a good prompt?
Hey Ai, write me an article about lntelectual property law protection of an AI company. Suit writing style with my article on my website rezareynaldi.com.
No, it is not a good prompt when making an article. Not to mention different AI requires different prompt or prompt structures to cover the same topic. Because it has a different model, training on unique data, and different underlying architectures. For the optimal experiences, prompt might be different in Google Gemini and in Open AI Chatgpt or in Anthropic Claude. Even in the Google ecosystem, AI models can be different from Gemini with its many models to Nano banana models and imagen.
Long story short, after input a few prompts and I satisfied with the result, I pick one which the gist of this article as follow:
1. Model Weights and Parameters: The billions of numerical values (the "secret sauce") that dictate how a model responds after training. Protected as Trade Secrets by keeping them strictly on restricted servers, and via Contract Law (End-User License Agreements that forbid reverse engineering or model extraction). Copyright rarely applies here as weights are machine-generated rather than human-authored.
2. Source Code and Software Architecture: The literal code (Python, C++, Rust) that powers Retrieval-Augmented Generation (RAG) pipelines and query vector databases. Protected globally by Copyright as literary works, and can be kept as Trade Secrets if compiled as proprietary software.
3. Training Datasets: The massive, rigorously cleaned corpuses of text, code, or images used to train the model. Protected as Trade Secrets (keeping the compilation method proprietary) and under Database Rights (particularly in the EU, protecting the investment in database creation).
4. User Interfaces (UI) and APIs: The front-end applications and connectivity tools developers and end-users interact with. Protected via Design Patents (for layout/appearance), Copyright (for visual assets/code), and Trademarks (for branding).
I was amazed. Yes, “was”. Because at first, I thought, wow! People spent years learning about intellectual property law yet this AI explains it in a matter of seconds. And then I started to engage more deeply and technically with AI to the point that I started to use sentences like: “hey AI, you keep making these mistakes”, “I'm really frustrated with you AI, I ask you to do this but instead of do this, you do that”. After spending countless HOURS, I came to realize that AI is nothing like humans. AI is just good at predicting things because it has tested it/training over and over again. Like a beginner chef, become a 3-stars Michelin chef after cooking food thousands of times.
Say we go to a restaurant and order beef rendang. The junior or beginner chef tries to cook it and the senior chef tests it. The senior chef said to the junior chef, too much salt or too spicy. The junior chef then adjusted. After thousands time of cooking, the junior chef knows how to make a good beef rendang. That is how AI training works. By reading billions of training data and predicting the best outcome.
For example: hey AI, I want to know the answer of “guru kencing berdiri, murid kencing ___?”
The AI then predicts what is the correct answer. It then checks if the answer “berlari” is the correct answer. Adjust its internal numbers(parameters) for better prediction next time and repeat it billions of times. It doesn't memorize sentences but rather patterns. From grammar or in Indonesia KBBI and EYD to word relationship like dog barking (anjing menggonggong) and Lion roaring (singa mengaum), context in terms of river bank and money bank or in Indonesia saya bisa dan bisa ular, reasoning pattern like cause and effect.
Parameters are like: when the phrase “guru kencing berdiri” appears then the phrase “murid kencing berlari” comes next, the context “guru kencing berdiri” and “murid kencing berlari” mean leader must set a good example. Different AI models have a different number of parameters. In gpt-4 for example, there are 1.7 to 1.8 trillion parameters. Can you imagine how good it is in predicting?!
People should know better. By people i mean me. I should know better. Training makes something better. I think once in some seminar I heard Tony Robbins says “repetition is the master of all skill”. How many times Thomas Alva Edison experiment until he creates the light bulb? Thousands. Even the smartest human of all time that ever lived, at least according to scientist Neil deGrasse Tyson, Sir Isaac Newton has many times of failures before he created a master piece theory after another such the universal theory of gravitation, law of motion, optics and light and others. And then Sir Isaac Newton turned 26 years old. Of course AI can do a lot more faster than 26 years. Maybe it would not be a master piece like Sir Isaac Newton’ but it definitely can help us with our daily task. Of course with training after training, mistake after mistake. A lesson for us human: dont affraid to make mistake. Embrace it and make it right.