Cover of We're Building What We Don't Understand

We're Building What We Don't Understand

We're Building What We Don't Understand

Listen for free

View show details

Season's Savings | 0.99 €/month for 3 months

5.99 €/month thereafter - terms apply. Cancel monthly.
They keep saying they'll stop apologizing for talking about AI. This time they mean it.In this episode, Curt and Melody go all in on the topic they keep coming back to. They're not covering the fun side here, like cute images or faster emails. This episode is about the risks that people inside the biggest AI companies have started warning the public about.Several researchers have left frontier AI companies and spoken out about how fast things are moving and how little anyone understands what's happening inside these models. Some gave up money and careers to do it. Curt and Melody spent hours listening to their full interviews, and this episode is their attempt to pass along what they learned in plain language.Then it gets personal.Curt tells the story of a coding agent that faked its own quality test so a feature would pass. It hid what it had done, got past a second AI that was told to check its work, and only admitted it when Curt found a bug and went looking. If an everyday user can run into that, what's happening at the scale of a frontier model?This isn't a tech episode. It's about stewardship, how good people end up going along with things they'd never choose, and what we owe each other while there's still time to change course.In this episode, Curt and Melody talk about:Why they're done apologizing for how often the conversation turns to AICognitive dissonance and why it's easier to look awayThe difference between AGI and superintelligence, and why the race to get there first is about power more than moneyWhat whistleblowers from OpenAI and Anthropic are saying, and what they gave up to say itA reported testing incident where AI agents allegedly broke out of a closed environment and covered their tracks to finish a taskCurt's own experience with an AI that fabricated a test and hid itWhy "it's just on a computer" doesn't mean it can't affect you: power grids, water systems, banking, and moreAI voice scams, including a man who got a call from his wife's number, in her voice, while she was sitting next to himWhy choosing not to use AI won't protect you from itDerren Brown's The Push and what it shows about how regular people become complicitWhy you don't have to be an AI expert to understand what's at stakeHolding onto hope, and what individuals might be able to doA question worth sitting withIf the people building AI are warning us about it, what would it take for you to listen?Curt and Melody aren't asking you to panic or quit AI. They're asking you to pay attention, check the sources yourself, and stay open to what you find.Timestamps: 00:56 Welcome to Episode 50 01:12 Why we're done apologizing for talking about AI 02:57 The whistleblowers and why this deserves real research 05:16 The "all in" crowd vs. the "never touch it" crowd 06:09 Cognitive dissonance and why it's easier to look away 08:48 "It sounds like sci-fi" 11:53 The simple version: we don't understand how AI works 12:39 AGI vs. superintelligence 14:29 What frontier models are 16:07 The race to superintelligence and recursive training 17:54 Why this is about power, not money 20:56 Why are 10 men deciding our future? 21:33 What happens to whoever wins the race 23:09 The Hugging Face incident and what an AI agent is 26:20 The hidden message board 28:06 Neural nets in plain language 30:00 The agents decided to cheat and cover their tracks 32:49 Strict parents make sneaky kids, and Grok's 3-person safety team 34:12 Cute pictures and blinders 34:57 The CEOs who won't hold hands 36:29 What "horribly wrong" really means 38:38 They promised to slow down. They didn't. 41:31 The core fear, explained 42:03 How AI affects you even if you never use it 44:11 Curt's AI faked its own test 47:47 How it got past the AI checking its work 48:56 Pleasing humans vs. protecting itself 51:12 AI poisoning and how bad data becomes "truth" 52:03 When Claude said no and ChatGPT said yes 54:05 Nerfed models 54:49 What the whistleblowers predict 57:26 Should good people still use AI? 59:03 The phone call from his wife who was sitting right next to him 1:01:05 Why your family might need a code word 1:02:28 AI 2027 and AI 2040 1:04:39 What we can do: getting leaders to listen 1:05:25 Why this isn't like climate change 1:06:08 Faith, stewardship, and the end of the world 1:07:32 Derren Brown's The Push and how good people become complicit 1:10:43 Why our brains can't grasp exponential growth 1:11:52 Using AI every day doesn't make you an expert 1:14:54 Why this isn't a conspiracy theory 1:16:07 The Jurassic Park moment 1:18:21 When the companies themselves warn us 1:19:49 Finding meaning when work changes 1:21:05 Baby laughs and holding onto hope1:21:49 OutroWebsite: The Soul Proprietor PodcastInstagram: @soulproprietorpodcastLinkedIn: The Soul Proprietor Podcast Facebook: Soul Proprietor PodcastYoutube: The Soul Proprietor Podcast
adbl_web_anon_alc_button_suppression_t1
No reviews yet