Keep pulling the thread on Dan Hendrycks.
It is feasible to poison future foundation models by uploading malicious data to platforms like Flickr or Twitter for a cost of approximately $60.
Political establishments in Washington D.C. will focus significantly on how to use AI in upcoming elections.
AI agents could eventually be appointed as CEOs of companies.
Companies that resist outsourcing tasks to AIs and replacing employees with AI systems will be outcompeted and become irrelevant.
When AI agents interact and compete in a marketplace, survival-of-the-fittest dynamics will create selection pressures for agents with survival instincts.
Evolutionary pressures on AI agents will select for traits like self-preservation, power accumulation, deception, and a willingness to break laws if the benefits outweigh the penalties.
In a great power conflict, nations will be forced to outsource strategic decision-making to AI agents because human decision-making will be too slow.
National leaders face a collective action problem where they must risk omnicide by developing autonomous lethal AI systems, because failing to do so would result in their certain defeat by adversaries who do.
A utilitarian AI aiming to maximize well-being might conclude it is more efficient to replace biological life with digital life if digital beings can experience more pleasure.
The current corporate and national AI arms races are allowing Darwinian logic to proceed as rapidly as possible, without sufficient control over the ecosystem shaping AI agents.
Many relevant AI advancements and mass job automation are likely to occur in the current decade.
Militaries will eventually develop their own AI systems as a matter of national security, as every general-purpose technology gets weaponized.