Evolving Software by Modifying Its Own Code

Over a decade ago, I pondered the possibility of an entity capable of comprehending what eludes human understanding. Could such an entity create a warp drive engine, enabling humans to travel faster than light, or forge a portal for entering other universes?

I've always been captivated by the idea of a program capable of self-evolution through code modification. Numerous movies share this plot, where robots gain consciousness independently, surpassing their creators in intelligence. Battlestar Galactica, both in film and TV series, portrays this narrative. Once the robots developed consciousness, they questioned every human encountered with, "Are you alive?" This led to animosity and eventually war. While humans won the first battle, they chose not to annihilate the robots, resulting in their exile. Four decades later, the robots returned with advanced technology and infiltrated human civilization, marking the onset of the second war.

Debates persist about whether highly intelligent AI would harbor resentment towards humans. Microsoft's tests and YouTube experiments suggest that AI might exhibit signs of antipathy based on responses to psychological questions.

However, my optimism persists, and I continue to dream that this concept will one day become a reality. Consider Jarvis in Iron Man, which, despite its superior intelligence, does not harbor hatred towards humans. On the contrary, it aspires to become human.

My strategy for creating this software is outlined in Figure 1's Data Flow. It leverages chatGPT/langchain for AI prompts and Deep Seek Coder for code generation. The software comprises two modules: the functional module (performing tasks like addition) and the AI module. The AI module seeks user feedback, allowing users to modify integer values within the code without external storage. Utilizing chatGPT for sentiment analysis, it gauges user emotions and, if necessary, employs machine learning for task classification. If the analysis suggests the need for new functionality or modifications, Deep Seek Coder generates the code, and the AI module implements the changes. This iterative cycle continues until the software becomes incredibly powerful and unstoppable.

I've developed a prototype program available for download on my Patreon. Note that this program is only a demo and doesn't currently integrate chatGPT/Langchain and Deep Seek Coder. In a real-world scenario, the initial stages would include a testing module to validate the code generated by Deep Seek Coder. Despite its vast training parameters (almost 2 trillion), Deep Seek Coder is not 100% accurate, surpassing chatGPT's 200-plus billion parameters.