It will be impossible to regulate and impossible to stop.
The code for AGI will not be some monolith of software architecture. The code will likely be simple. This means that someone in their basement could build it. The steps to get there are challenging, though. A single person could have developed the transformer architecture. A single person could have used 4-bit quantization and developed an AI that is just as good as ChatGPT and have it run on their local machine.
The difficulties are figuring out the best 'needle in the haystack' to solve the problem. This requires research, and this process happens much faster if you have more people working on it. For years, many people did not put any energy into AI systems because the hardware was not here yet. The hardware is here now. The cat is out of the bag.
The code for AGI will not be some monolith of software architecture. The code will likely be simple. This means that someone in their basement could build it. The steps to get there are challenging, though. A single person could have developed the transformer architecture. A single person could have used 4-bit quantization and developed an AI that is just as good as ChatGPT and have it run on their local machine.
The difficulties are figuring out the best 'needle in the haystack' to solve the problem. This requires research, and this process happens much faster if you have more people working on it. For years, many people did not put any energy into AI systems because the hardware was not here yet. The hardware is here now. The cat is out of the bag.