The AI Dilemma, RoboCat and How to do Great Work – Live and Learn #21
Welcome to this edition of Live and Learn. This time with a video explaining how harmful generative AI is going to be, a better way for AI to learn real-world tasks and an excellent article by Paul Graham. Enjoy.
✨ Quote ✨
While superintelligence seems far off now, we believe it could arrive this decade.
— OpenAI - (source)
🖇️ Links 🖇️
The A.I. Dilemma by Tristan Harris and Aza Raskin. A great video describing why AI, even only as powerful as we have it right now is a great danger to our societies. The flood of generative content will break our systems. AI is the equivalent of a digital nuke. It can be used to manipulate millions of people, creating fake viral videos of events that never happened. Our social media constructs are not equipped to handle this. The rise of bots capable of pretending to be real humans, generating completely artificial, but highly believable stories, designed to manipulate and deceive, will break our existing information infrastructure.
Why AI is Harder than We Think by Melanie Mitchell. This paper dissects four fallacies, which make us overestimate how soon superintelligence is going to arrive. It's interesting to see how far apart the opinions of experts are on this topic. Melanie Mitchell, the author of this paper also took part in this debate, where the two sides argue over whether AI is an extinction-level risk.
How to do Great Work by Paul Graham. Great Work happens when you follow your curiosity no matter what. Things that interest you, if pursued far enough, lead to great work. Like any work of Paul Graham, this one is worth reading.
RoboCat by DeepMind. This blog post and paper describe an approach for creating self-improving robotic agents. DeepMind is figuring out how to build robotic arms that can learn new tasks on their own, which will help in eventually building a general-purpose robot, that can do everything a human could.
🌌 Midjourney 🌌
🎶 Song 🎶
Barfuß am Klavier by AnnenMayKantereit
I hope you found this newsletter useful, beautiful, or even both!
Have ideas for improving it? Please let me know.