Notes
The initial solution:
- Enhanced oversight
- Liability
Extensive tool AI under human control should be the goal.
General AI engineering direction:
- Understanding what I want is what the AI wants as well, these are the only two people who are relevant and who the AI should care about.
- Act always and only in accordance with our wants.
- Genuinely care, doesn't just act like it cares
Hardware based security mechanisms may work… on chip hardware features.
US Chips Act and Export Control Regime.
Humans should be deeply involved, and the AI's scope of action should be severely limited.
“For example, governments able to tax AI companies could invest a fraction of receipts into a sovereign wealth fund that holds stock in the companies, and pays dividends to the populace.”
This is already happening with top AI companies, as well as SpaceX now, so that’s already happening through the NASDAQ, pension funds and forms of passive investing, and finance experts say the bubble will burst so everyone will actually LOSE money…
Widely available AI assistants that put users interests above corporate providers.
We need to control what AI and why AI. Not just to preserve our mental and physical abilities as humans, but also realize and understand what we can genuinely do better than any AI can or will be able to in the future.
Review
8/10.
Phenomenal book, a must-read for every technical expert working on AI technologies.
Uses a lot of citations to make points that clearly emphasize the current direction AI is going in, and where it has the potential to make huge impacts on the world that we simply don't need and aren't ready for in today's time.
Also covers potential solutions and directions.