There's no reason to suppose that we won't continue to make progress in AI, and eventually reach AGI. But I don't think it'll be by adding 10 times as much text to the training set for chat GPT or going from 100 billion parameters to a trillion parameters. The universe just doesn't have enough data and the earth Doesn't have enough physical resources to build a machine big enough to achieve AGI using these simple circuit optimization techniques. We see this with computer vision systems where you know, you need millions of examples of a giraffe before it gets good at recognizing giraffes.
The Sunday Times’ tech correspondent Danny Fortson brings on Stuart Russell, professor at UC Berkeley and one of the world’s leading experts on artificial intelligence (AI), to talk about working in the field for decades (4:00), AI’s Sputnik moment (7:45), why these programmes aren’t very good at learning (13:00), trying to inoculating ourselves against the idea that software is sentient (15:00), why super intelligence will require more breakthroughs (17:20), autonomous weapons (26:15), getting politicians to regulate AI in warfare (30:30), building systems to control intelligent machines (36:20), the self-driving car example (39:45), how he figured out how to beat AlphaGo (43:45), the paper clip example (49:50), and the first AI programme he wrote as a 13-year-old. (55:45).
Hosted on Acast. See acast.com/privacy for more information.