AGI Definition, So You Don't Get Tricked Anymore

Many people are confused about what Artificial General Intelligence (AGI) truly is. Recently, OpenAI claimed they developed AGI again, but the concept remains unclear. To better understand AGI, it's essential to distinguish it from current AI systems and the underlying technology.

Determinism and HTTP request-response architecture, which are prevalent in modern computing, do not support the random or non-specific way AGI should operate. Current AI algorithms rely on precise, deterministic processes where the same input always yields the same output. This is due to the binary nature of hardware, which can only handle 0s and 1s.

AI systems today use HTTP and do not store tokens (pieces of context) between requests. Instead, they cache data but do not use it to retrain or improve their models. This means the AI is merely processing and caching data rather than truly understanding or reasoning about it. The phenomenon of AI "hallucinations" (inaccurate or nonsensical outputs) arises from the system's inability to reconsider or self-correct, as its computations are fixed and unchangeable.

In theory, AGI should be able to process data without caching requests or responses, operate without predetermined random token generation, and exhibit a non-stop, self-improving system akin to a human brain. AGI should be able to engage with users in real-time, without the need for transferring entire neural networks or having predefined response sizes. This hypothetical AGI would be capable of growing its size and adapting to new information over time, potentially surpassing human capabilities in specific tasks.

While some may argue that AGI is already here, in practice, current AI systems are simply advanced forms of sophisticated autocomplete based on human language patterns. To achieve AGI, significant advancements in hardware, software, and our understanding of intelligence are required. This could involve developing a physical, self-improving neural network that can operate indefinitely and learn from its environment continuously. The prospects are exciting, but the path to AGI is complex, and significant challenges lie ahead.