00:00Hundreds of people just took to the street to protest AI's next big breakthrough.
00:05And I don't blame them.
00:06Things are looking a little unsettling.
00:08Companies like OpenAI and Anthropic are pouring billions of dollars
00:12into building so-called recursive AI, or AI that can essentially build itself.
00:17One campaigner who went on a seven-day hunger strike
00:20outside the offices of Google to protest AI's development
00:23pointed to that new field.
00:26We believe that building AI that can automate AI research
00:29and that can self-improve could be like a danger to the human race
00:32and potentially lead to human extinction.
00:34Not only me and other researchers saying this,
00:37it's the lab sales themselves say that the risk is real.
00:41He's right.
00:41Anthropic recently published this blog,
00:44noting that most of its code is now generated by clod.
00:48And humans have become the bottleneck in the development process.
00:52Some campaigners and researchers want to pause
00:55and discuss the implications of that
00:57before allowing the development of AI to go any further.
01:00But at this point, it's not clear if that's possible.
01:03AI startup founder Tim Rocktashel suggests that ship has sailed.
01:08His company, Recursive Superintelligence,
01:10is completely focused on creating self-improving AI and using it.
01:15He assured me that all the development is done in a sandboxed environment.
01:20In other words, a self-evolving AI model
01:22wouldn't escape onto the internet and wreak havoc.
01:25Instead, he says it acts like a junior researcher
01:29who must convince its human supervisors of a sound scientific result.
01:34Now, the analogy sounds compelling,
01:36but I'd prefer more concrete evidence that this is all completely safe.
01:40Self-improving AI isn't like a switch you can just turn off like the protesters want.
01:46It's more like a dimmer.
01:47Its use is increasing little by little inside the world's biggest AI labs.
01:52Self-improving AI
01:53Self-improving AI
Comments