There are many parallels between how human beings and LLMs operate, but there might be a crucial difference — LLMs could potentially never die.
One of the godfathers of AI and Nobel Laureate Geoffrey Hinton says that LLMs are essentially immortal. He says that the LLM’s entire personality lies in its weights, and if similar hardware were created years later and the weights run on them, the LLM would essentially be reborn. He says this is different from the human anatomy where the weights and the hardware are both in the brain, and are destroyed when the brain is destroyed.
“These things are immortal — these large chat bots are immortal,” he said in a talk. “If you keep a copy of the weights somewhere, you can destroy all of the hardware they were using, build more hardware later, put the same weights on that hardware and they’ve come back to life. The very same thing has come back to life, the very same being,” he added.
Hinton however said that humans are fundamentally built differently. “But I got interested in the fact to achieve that kind of immortality, we have to have the hardware do exactly what we tell it to do with the program. And so I decided to explore what would happen if we abandoned that principle of separating the software from the hardware. And we had things like our brains in which there’s no distinction, the connection strengths is in your brain, and no use to anybody else — they’ve got a different brain with neurons, with different properties connected in different ways. And your connection strengths have no interest to them,” he added.
“This dream (some people have that) they’re gonna upload themselves to a computer is just nonsense,” Hinton said. “The connection strengths you have that make you, you are intimately related to the particular neurons you have that make you. You can’t upload your weights and have them run on some other hardware,” he added.
The neural networks that power LLMs are modelled on the neurons inside the human brain. But unlike in the human brain where where the “hardware” and the neurons themselves are tightly coupled, the “neurons”, or weights of a model, are independent of the hardware. These weights can be downloaded and run on a different machine, and the LLM is essentially reborn in that machine. But in the human brain, the hardware and the weights are essentially placed together, and work in unison. As because they’re inextricably linked to each other, Hinton argues that it would be impossible for humans to be reborn — at least in a manner analogous to how LLMs can be.