Large language models and general "artificial intelligence"
vendredi 19 mai 2023 à 00:32Confusion of large language models with general "artificial intelligence" is causing panic.
Artificial general intelligence may someday be developed, and if so it could pose many sorts of dangers, I recommend _A Fire upon the Deep, for an example, and Accelerando. And many other stories, with a wide variety of outcomes.
I agree that these systems can be dangerous. But they are dangerous in the short term in ways that hardly resemble the disaster scenarios. Indeed, artificial general intelligence is unnecessary for digital systems to bring disaster. Imagine payment systems that track all purchases, and transportation systems that track people's travel. Those already exist, and if we don't want our countries to be like China, we need to abolish them.
ChatGPT is nothing like that. It can't make coherent plans, let alone carry them out. It can generate text that looks like a plan until you study it carefully. The problems it causes come from side effects.