Seonglae ChoMeta-algorithms that complete themselves through data compression
Humans imagined AI as a unified, mechanically synchronised massive intelligence, but artificial intelligence, which is much easier to replicate and modify than human brain, is created with far more diverse tastes and sizes tailored to specific situations. This diversity and AI society mediate human society and are ambiguously superpositioned, enabling collaboration on a larger scale.
AI Notion
AI Usages
How AI is Understanding the World
Threshold driven development
Software 1.0, Software 2.0, Software 3.0
Software 1.0 is built with code, while Software 2.0 is built with weights. Traditional Software 1.0 can automate everything that can be specified. AI can automate everything that can be verified. Software 3.0 uses prompts to enable LLMs as programmable neural networks. We can add one analogy on this: LLM is a new operating system that can be accessible through CLI, prompt.
Partial autonomy app such as Cursor.so integrate llm next to traditional interface. Application specific GUI with AIs under the hood with autonomy slider. You can tune the amount of autonomy to give up. Perplexity AI can be also explained in same way. AIs are doing the generation and human is doing verification. However, reading is boring and time-consuming while GUI is a highway to your brain for auditing system and visual representation. Conclusively, we need to on the lease AI agents to make it more successful.
