You are right, the concept of AGI is a business for companies, and while they talk about it as something achievable through LLMs, that’s not the case. For example, the O1 systems are multiple instances of GPT-4 set up in a loop, asking and answering each other to reach an acceptable conclusion, but that’s not real reasoning. (You can try asking O1 who it is, and it will tell you it’s GPT-4.) No matter how much training you give an LLM, it won’t truly think or speak better, although it can reflect certain intelligent behaviors. Regarding the human brain, there are obviously many unknowns, but the things we do know can be replicated—emotions, memories, etc.—and a system can be synthesized that is closer, if not a complete AGI.
5 Likes