0:00 The data underneath AI is quietly rotting.0:02 A new thread shows as little as one percent synthetic, AI-generated data can poison a training set, causing models to lose detail and make weaker judgments.0:14 This quiet crisis of data integrity is the hidden context for why the biggest players are suddenly changing how their AIs work.0:22 The other big moves this week all circle this same theme of control.0:27 California just drew a line in the sand for AI vendors.0:31 Governor Gavin Newsom signed an executive order requiring any company to prove its models are safe from bias and protect civil rights before getting a state contract.0:43 It's the first major procurement wall in the U.S.0:46 At the same time, the brokerage platform Public just launched AI Agents that automate entire portfolio strategies.0:54 You no longer place orders; you state an intent, and the agent executes, monitors, and adjusts.1:01 And on the security front, Palo Alto Networks CEO Nikesh Arora just warned that autonomous AI agents could shrink the time for a sophisticated cyberattack down to just twenty-five minutes, a fundamental shift in the threat landscape.1:18 But let's go back to that rotting data.1:20 Because Microsoft just revealed its strategy for dealing with an unreliable world, and it's a major tell.1:28 They've revamped their internal AI research tools to integrate models from both OpenAI and Anthropic.1:35 Here's the turn.1:36 They introduced a new layer called "Critique," where they use Anthropic’s Claude model specifically to review and find flaws in answers generated by OpenAI’s models.1:48 One AI is now fact-checking another.1:50 They claim this multi-model approach improved research quality by almost fourteen percent.1:57 That’s not a small number.1:59 This isn't just a new feature.2:01 It’s an admission.2:02 The problem of "model collapse"—where AIs trained on other AIs' outputs get progressively dumber—is no longer a theoretical risk.2:11 It's an active engineering problem inside Microsoft.2:15 They can no longer trust a single model’s output on its own.2:19 So they're building an immune system.2:22 Charles Lamanna, a Microsoft EVP, said it plainly: "It’s becoming very clear to us that there will be many models." The future isn't one god-model.2:32 It's a council of them, watching each other.2:35 The era of treating large language models as infallible black boxes is officially over.2:41 The new race isn't just for more capability.2:44 It's for verifiable reliability, whether that's through government contracts, automated market agents, or AIs policing other AIs.
