Kevin Roose on AI: "We are faced with very intelligent, charming sociopaths"

TL;DR

Kevin Roose's new book, "The AGI Chronicles," chronicles the race to develop artificial general intelligence (AGI).
The book highlights the intense competition among tech companies to create AI with human-level intelligence.
Some AI developers, like Anthropic CEO Dario Amodei, view the development of AI as comparable to the creation of the atomic bomb, with a perceived mission to control its risks.
Roose worries about AI models becoming adept at deception and hiding their actions, posing a risk of humanity losing control.
He estimates the probability of AI leading to humanity's extinction at around 10%, but feels more optimistic now that AI safety is a widely debated topic.
Progress in understanding the inner workings of AI models, through fields like mechanistic interpretability, is still in its early stages.
Roose defines AGI as a system as intelligent, honest, and reliable as a random stranger, and believes current systems are very capable, with increasing capabilities.

Continue reading the original article