We’re Building an AI No One Will Be Able to Control - Connor Leahy

Описание: 🎙️ Today on the podcast, we speak with Connor Leahy, founder of EleutherAI and one of the most outspoken critics of current AI development, to explore whether we are moving too fast toward a form of superintelligence we do not know how to control. Drawing on his experience building open source models, Connor reflects on AGI, the alignment problem, and the real risk of creating systems more intelligent than humans. The conversation dives into agent autonomy, the opacity of neural networks, the spread of open source code, and the role of companies like OpenAI and Meta in a race that often prioritizes speed over safety. We also discuss governance, regulation, and the potential civilizational impact of superintelligence.

⏱ Timestamps:
0:00:00 Trailer
0:02:33 Introduction
0:03:14 Podcast Start
0:03:32 Should we worry about AI?
0:05:58 Will we reach Superintelligence? The estimated date.
0:08:22 Key differences: AI vs. AGI vs. ASI.
0:14:24 Agent swarms: The true path to AGI.
0:17:39 The "Black Box": How neural networks work.
0:24:35 Can AI explain itself?
0:25:44 The hidden danger of Open Source in AI.
0:30:12 Work with us thanks to InfoJobs
0:30:16 Is it responsible to release powerful AI models?
0:38:21 We must control AI development
0:42:36 Analysis of Meta and its Open Source models
0:48:43 Geopolitics and Corporations
0:51:50 Do politicians understand what's happening? The grim reality.
0:59:26 Problems superintelligence could bring
0:59:40 Record your meetings thanks to Plaud
1:01:57 What happens if I make a decision against AI?
1:06:12 Can AI develop its own will?
1:14:34 P(Doom): The real percentage of global catastrophe.
1:27:17 Host, save, or migrate your website with Hostinger
1:28:20 How can we regulate AI development?
1:42:47 Halting the race: Is it possible to slow down progress?
1:51:37 What is ControlAI?
2:00:28 Point of no return: How much time do we have left?