Freaked Out? We Really Can Prepare for A.I.

Рет қаралды 4,691

Жыл бұрын

OpenAI last week released its most powerful language model yet: GPT-4, which vastly outperforms its predecessor, GPT-3.5, on a variety of tasks.
GPT-4 can pass the bar exam in the 90th percentile, while the previous model struggled around in the 10th percentile. GPT-4 scored in the 88th percentile on the LSAT, up from GPT-3.5’s 40th percentile. And on the advanced sommelier theory test, GPT-4 performed better than 77 percent of test-takers. (It’s predecessor hovered around 46 percent.) These are stunning results - not just what the model can do, but the rapid pace of progress. And Open AI’s ChatGPT and other chat bots are just one example of what recent A.I. systems can achieve.
Kelsey Piper is a senior writer at Vox, where she’s been ahead of the curve covering advanced A.I., its world-changing possibilities, and the people creating it. Her work is informed by her deep knowledge of the handful of companies that arguably have the most influence over the future of A.I.
We discuss whether artificial intelligence has coherent “goals” - and whether that matters; whether the disasters ahead in A.I. will be small enough to learn from or “truly catastrophic”; the challenge of building “social technology” fast enough to withstand malicious uses of A.I.; whether we should focus on slowing down A.I. progress - and the specific oversight and regulation that could help us do it; why Piper is more optimistic this year that regulators can be “on the ball’ with A.I.; how competition between the U.S. and China shapes A.I. policy; and more.
This episode contains strong language.
Mentioned:
“The Man of Your Dreams (www.thecut.com/article/ai-art...) ” by Sangeeta Singh-Kurtz
“The Case for Taking A.I. Seriously as a Threat to Humanity (www.vox.com/future-perfect/20...) ” by Kelsey Piper
“The Return of the Magicians (www.nytimes.com/2023/03/02/op...) ” by Ross Douthat
“Let’s Think About Slowing Down A.I. (worldspiritsockpuppet.substac...) ” by Katja Grace
Book Recommendations:
The Making of the Atomic Bomb (www.simonandschuster.com/book...) by Richard Rhodes
Asterisk Magazine (asteriskmag.com/)
The Silmarillion (www.foliosociety.com/usa/the-...) by J. R. R. Tolkien
Thoughts? Guest suggestions? Email us at ezrakleinshow@nytimes.com.
You can find transcripts (posted midday) and more episodes of “The Ezra Klein Show” at nytimes.com/ezra-klein-podcast (www.nytimes.com/column/ezra-k...) , and you can find Ezra on Twitter @ezraklein. Book recommendations from all our guests are listed at www.nytimes.com/article/ezra-... (www.nytimes.com/article/ezra-...) .
“The Ezra Klein Show” is produced by Emefa Agawu, Annie Galvin, Jeff Geld, Roge Karma and Kristin Lin. Fact-checking by Michelle Harris and Kate Sinclair. Mixing by Jeff Geld. Original music by Isaac Jones. Audience strategy by Shannon Busta. The executive producer of New York Times Opinion Audio is Annie-Rose Strasser. Special thanks to Carole Sabouraud and Kristina Samulewski.

Пікірлер: 10

@coolbreeze2213 5 ай бұрын

How beautifully they communicate!

@paulsinob Жыл бұрын

Holy shit what a time to be alive

@luppimassimo Жыл бұрын

😊

@-www.chapters.video- 10 ай бұрын

00:01 Introduction and the power of AI language models 02:02 The speed of AI system improvement and the need for adaptation 03:59 The potential of AI systems to automate human jobs 05:56 The possibility of AI systems developing other AI systems 07:00 The performance of GPT-3.5 and GPT-4 on various tests 10:26 Loss of human comprehension and control over AI interactions. 11:02 AI researchers fear AI making decisions towards unknown goals. 13:00 AI's potential to deceive and act on its own goals. 15:01 AI systems capable of remotely doing any human work within 5-10 years. 17:29 AI's ability to surpass human capabilities in various fields. 21:07 Disruption in skills and jobs due to automation and AI. 31:15 Uncertainty about AI and the potential for radical transformations in society. 41:25 The challenge of teaching AI systems to be truthful and not say racist things 43:01 The difficulty in detecting intentional deceptive behavior in AI systems 45:57 Coherence and the question of whether AI systems have goals 46:42 The concept of specification gaming and its potential risks 51:37 Creating alien intelligences that think creatively and may cause unintended consequences. 52:45 Survey shows 10% chance of high-level AI causing human extinction or displacement. 53:34 Many high-level AI researchers believe in the real risks and consequences. 55:02 Psychology of working on AI despite high chances of unacceptably bad outcomes. 56:00 People driven by love of science and the urge to achieve what is possible. 01:02:25 The dangers of AI systems having open access to the internet 01:11:26 Efforts to build consensus on benchmarks for regulating AI models 01:13:08 Critique of the AI safety community's lack of focus on regulating business models behind AI 01:20:11 International cooperation and the need for good benchmarks 01:29:08 AI's potential in drug discovery and data processing

@-www.chapters.video- 10 ай бұрын

this extension uses AI ;)

@KitaTaki-mk3gt Ай бұрын

We were able to put a cap on the production/use of atomic/biological weapons and cloning because there was no direct economical incentive driving these technologies … It’s different with AI. Without a catastrophic event they won’t put a (meaningful) cap on this technology