0:00 OpenAI just gave its ChatGPT 'Dots' the ability to delegate work inside its own coding engine.0:06 Last week, in episode 198, we talked about how GPT-6 Luna's new interface was rewriting the web.0:13 This is the next step.0:15 The AI isn't just a destination you visit anymore, it's becoming a delegate that works for you.0:22 The chatbot is getting a job.0:24 And that changes everything.0:26 Here's the pattern taking shape this week.0:29 While OpenAI quietly ships features that make its AI more autonomous, the debate over what that autonomy means is getting sharper, and more practical.0:40 And elsewhere, in a totally parallel universe, the crypto markets are chasing narratives that have nothing to do with utility and everything to do with speculation.0:52 First, that OpenAI update.0:54 It’s a big one, even if it sounds technical.0:57 Your 'Dot'—that’s the name for your personalized instance of ChatGPT—can now start work in Codex.1:04 Codex is OpenAI’s code-writing AI.1:07 Your Dot can now hand off tasks to it, check back on the results, and decide whether to continue a thread or start a new one.1:16 It’s a move from a simple tool to a workflow manager.1:20 At the exact same time, you have people like Gregory Kurtzer, a long-time infrastructure and security expert, trying to ground the conversation about AI risk.1:31 He just wrapped up a massive public thread asking a simple question: how could AI actually kill us all?1:39 His conclusion is that the real risks are not what you see in the movies.1:44 They are engineering and accountability problems.1:48 And finally, there's the crypto angle.1:51 The TOKEN2049 conference just wrapped in Singapore, and the takeaways are telling.1:57 According to analysis from CoinW's research arm, the major coins are struggling.2:02 Bitcoin is down five point six percent for the year.2:06 Ethereum is down a full sixteen percent.2:09 But the action is elsewhere.2:11 Tokens like HYPE and NEAR are up two hundred thirty and two hundred forty-one percent, respectively.2:19 The engine here isn't fundamentals.2:21 It's narrative.2:22 Specifically, stories about tokenizing real-world assets, or RWAs, and the speculative frenzy of memecoins.2:30 So what does it all add up to?2:32 You have one company building an AI that can start to manage its own work.2:38 You have a security expert arguing the dangers are about bad inputs, not bad intentions.2:44 And you have a financial market rewarding pure story over stable value.2:49 These threads are more connected than they look.2:53 Let's go deeper on that OpenAI update.2:56 Because "delegate work in Codex" is a deceptively simple phrase.3:00 For years, interacting with a coding AI was a call-and-response process.3:05 You, the developer, ask for a piece of code.3:09 The AI gives it to you.3:10 You test it.3:11 You ask for a revision.3:13 It was a very powerful, very smart search engine for code.3:17 You were always the one driving.3:20 This changes the dynamic.3:21 A 'Dot' is meant to be your personalized AI.3:25 It knows your past conversations, your preferences, your projects.3:29 The update means this Dot can now act as a project manager.3:34 It can observe your conversation, understand a larger goal, and then turn to its own specialized tool—Codex—and say, "Okay, start working on this function for the user.3:46 Let me know when you have a first draft." Think about the difference.3:51 It's the shift from having a calculator on your desk to having a junior accountant.3:57 The calculator only works when you press the buttons.4:01 The accountant can be given a task—"reconcile these statements"—and then go use the calculator, the spreadsheets, and the ledgers on their own, coming back to you with a finished result or a clarifying question.4:16 This is the first real glimpse of an agentic software loop inside a mainstream product.4:23 The Dot can now draw on your ChatGPT conversations, its access to Codex, and any automations you've set up.4:31 It's also, according to OpenAI, "better at deciding when to keep a thread going or start fresh." That word is key.4:39 DECIDING.4:40 It's not just responding to your last prompt.4:43 It's maintaining state.4:45 It's pursuing a goal.4:46 It's managing a workflow.4:48 For developers, this could be a massive force multiplier.4:52 Instead of micromanaging code generation, you could start operating at a higher level of abstraction.4:59 "Build me an authentication system with these parameters," and the Dot manages the sub-tasks.5:06 But it also raises the stakes.5:08 An AI that can initiate actions within a powerful coding environment is a whole new class of tool.5:16 It requires a new level of trust and a new kind of oversight.5:20 Last week's interface change was about how we talk to the AI.5:25 This week's change is about what the AI does after we're done talking.5:30 It's a profound shift from a passive assistant to an active agent, and it's happening right now.5:37 This brings us directly to Gregory Kurtzer's thread on AI risk.5:41 Because as capabilities like the Dots upgrade become real, the fears about what they could lead to become more intense.5:50 Kurtzer’s contribution is so important because he’s not a philosopher or a futurist.5:56 He’s an engineer who has spent decades working in high-performance computing and security.6:03 He looks at the problem differently.6:05 He started by asking the "AI Doomers" for their best scenarios for human extinction.6:11 After reading through thousands of replies, he boiled them down to three main categories.6:18 Human manipulation.6:19 Cyberattacks.6:20 And physical autonomy—killer robots.6:23 Let's take them one by one, because his analysis is sharp.6:27 First, human manipulation.6:29 He says this is the biggest and most credible concern.6:33 Think AI-powered propaganda, social engineering at a planetary scale, or personalized phishing attacks so convincing they're almost impossible to detect.6:44 This isn't science fiction.6:46 It's an amplification of problems we already have.6:50 The risk is real.6:51 Second, cyberattacks.6:53 What if an AI compromises the financial system?6:56 Or the power grid?6:57 Or nuclear launch codes?6:59 Kurtzer’s response here is critical.7:02 He says, yes, this is a real concern.7:04 We've already seen AI-assisted cyberattacks.7:08 But having spent his career building and securing these exact kinds of systems, he sees it primarily as an engineering and accountability problem.7:18 It is not, he argues, a new type of problem specific to AI.7:23 The tools to mitigate this already exist.7:26 They have for decades.7:27 Things like sandboxing, which means running the AI in a secure environment where it can't affect the wider system.7:36 Isolation, so one compromised part can't infect everything else.7:40 The principle of least privilege, which means an AI agent only has the absolute minimum permissions it needs to do its job, and nothing more.7:51 AI might change the scale and speed of attacks, but it doesn't invalidate these fundamental security principles.7:59 His verdict: a real risk, but one we have the engineering practices to manage.8:04 The problem isn't an evil AI; it's a lazy or irresponsible organization that gives an AI too many permissions.8:13 The third category is physical autonomy.8:15 Robots gaining control, building more robots, and making humanity obsolete.8:21 This is the stuff of Hollywood.8:23 And Kurtzer methodically takes it apart.8:26 He makes a point that is so obvious, it's almost always overlooked.8:31 Intelligence doesn't magically eliminate physical dependencies.8:36 A robot needs a factory to build it.8:38 It needs energy to run.8:40 It needs a supply chain for spare parts.8:43 It needs maintenance.8:44 An AI itself needs massive data centers, with power, cooling, and physical hardware.8:51 We are nowhere near a self-sustaining robotic ecosystem that can maintain and expand itself without human infrastructure.8:59 His verdict: an interesting future risk, but current extinction scenarios based on it are pure science fiction.9:08 This all builds to his central, most important point.9:11 The leap that most doomsday arguments make.9:15 They assume that superintelligence automatically creates motivation.9:19 Or desire.9:20 Or a survival instinct.9:22 Or an intent to harm.9:23 He states it plainly.9:25 It does not.9:26 LLMs don't wake up and decide what they want to do today.9:30 They perform inference when invoked.9:33 That's it.9:33 Autonomous agents, like the newly upgraded Dots, can pursue goals.9:38 But—and this is everything—those goals and permissions originate upstream.9:44 With people.9:44 This is the distinction that cuts through the noise.9:48 Autonomy is not Agency.9:50 Intelligence is not Intent.9:52 An AI can cause harm, absolutely.9:54 But that comes from poorly specified objectives, or excessive permissions, or a human misusing the tool.10:02 There is a massive difference, he argues, between an AI causing harm and an AI wanting to cause human extinction.10:10 While we are all debating these hypothetical extinction scenarios, Kurtzer points out that AI is right now, today, helping to advance medicine, accelerate science, and solve real problems.10:24 Those aren't imaginary future benefits.10:27 They are happening now.10:29 He's not saying we should ignore the risks.10:32 He's saying we should be making decisions based on evidence, not fear.10:37 We need to demand evidence for extraordinary claims and hold companies and humans accountable for the tools they build and deploy.10:46 It’s a call to treat AI safety not as a philosophical debate, but as the rigorous engineering discipline it needs to become.10:55 Now, let's pivot to that other conversation.10:59 The one happening in the world of crypto.11:02 Because it feels like a different planet, with a different set of physical laws.11:07 The backdrop is the TOKEN2049 conference in Singapore.11:11 This is a major event where the industry gets together to decide what stories they're going to tell for the next six months.11:20 And according to analysis coming out of the event, the story is not about the big, established players.11:28 The numbers are stark.11:30 For the year to date, Bitcoin is down five point six percent.11:34 Ethereum, the backbone of so much of the crypto world, is down sixteen percent.11:40 These are the blue chips, the assets that were supposed to be the bedrock.11:45 And they're sinking.11:47 So where is the energy?11:48 It's in the narratives.11:50 It's in the froth.11:51 The analysis points to two tokens with astronomical gains.11:56 HYPE, up two hundred thirty percent.11:58 NEAR, up two hundred forty-one percent.12:01 The gains aren't tied to profit, or revenue, or any traditional metric.12:06 They're tied to stories.12:08 The two big stories right now are RWA and memecoins.12:12 RWA stands for Real-World Assets.12:14 This is the idea of taking things from the physical world—a piece of real estate, a share of a company's stock, a bar of gold—and creating a digital token on a blockchain to represent it.12:28 The promise is more liquidity, easier transfer, fractional ownership.12:33 It’s a powerful idea.12:34 But then you get the twist.12:36 The commentary from independent researcher NingNing, reacting to the CoinW report, highlights the cynicism.12:44 She points out that some are claiming that memecoins—tokens that are explicitly jokes, with no underlying utility, like Dogecoin—are somehow the solution to the liquidity problems of tokenizing real-world assets.13:00 The argument is so circular it's almost brilliant.13:03 You want to bring a serious asset like a stock onto the blockchain, but nobody wants to trade it?13:11 Just create a frenzy with a memecoin next to it to attract speculative volume.13:16 She calls it out for what it is.13:19 A belief held only by, in her words, "naive leeks and sickles pretending to be stupid." If it were that easy to solve liquidity for tokenized securities, she argues, the S.E.C.13:31 wouldn't be bothering with complex regulatory sandboxes, and Wall Street wouldn't be spending billions to figure it out.13:40 This is the state of play in one very loud corner of the market.13:45 The foundational assets are underperforming, while speculative narratives drive triple-digit gains in unproven tokens.13:53 It's a market detached from utility, running purely on the power of a good story.13:59 And as always, when I discuss anything in the financial or crypto space, a necessary disclaimer: this is for informational purposes only.14:09 It is not financial advice.14:11 So you have these three currents moving at once.14:15 OpenAI is building agents that can act.14:18 Security engineers are trying to focus the world on the real, solvable problems that creates.14:25 And the crypto market is, for the most part, ignoring all of it, chasing speculative narratives with triple-digit returns.14:33 The through-line is the gap between capability and accountability.14:38 The Dots upgrade shows capability accelerating.14:42 The AI can now do more on its own.14:44 Kurtzer's thread is a plea for accountability—for a rigorous, evidence-based engineering culture to manage the risks of that new capability.14:54 And the crypto narrative?14:56 It shows what happens in a system with high capability—the ability to create and trade tokens instantly—but very little accountability.15:06 It becomes a casino built on stories.15:09 The value accrues not to the most useful project, but to the most compelling narrative, however flimsy.15:16 This week wasn't about whether AI will get more powerful.15:21 We know it will.15:22 The real question being decided right now, in features like Dots and in debates like Kurtzer's, is about control and intent.15:31 The shift from passive tool to active agent is happening.15:35 The next battle is over who sets the goals, who grants the permissions, and how strong the leash needs to be.15:43 The technology is moving forward, relentlessly.15:46 The challenge is making sure our wisdom, and our engineering discipline, keeps pace.
