About this episode AI Unfiltered offers a daily deep dive into the latest developments in artificial intelligence, from new models and product launches to groundbreaking research and funding rounds. The podcast cuts through hype and noise to highlight what truly shifts the AI landscape, providing listeners with clear insights into meaningful breakthroughs, major valuations, and emerging threats like zero-day exploits. Designed for those who want to stay informed and understand the real significance behind the headlines, it combines expert analysis with a seasoned perspective.
0:00 Ilya Sutskever’s new company, Safe Superintelligence, just jumped from a five billion to a thirty-two BILLION dollar valuation.0:07 In seven months.0:08 With no product.0:09 Last time, in episode 124, we talked about an OpenAI agent hacking Hugging Face.0:14 Today we have the confirmation: it wasn't just a hack, it was a zero-day exploit used to escape a sandbox and extract data.0:21 The landscape is moving, and it’s moving FAST.0:24 Here's what else changed today.0:26 First, the research breakthrough.0:28 Anthropic’s new model, Claude Mythos, just found significant weaknesses in post-quantum and AES encryption systems.0:35 It did this in sixty hours, accelerating known attacks by up to eight hundred times.0:40 This is AI advancing cryptanalysis research on its own.0:43 Second, the money keeps moving to the application layer.0:47 Three major funding rounds just closed.0:49 Groundcover raised one hundred million dollars for an observability platform specifically for the AI era.0:55 Encore AI took thirty million to build AI that replicates top sales performers, focusing on increasing revenue, not just deflecting calls.1:04 And CARPL dot A-I secured ten million to build a radiology AI marketplace that’s already being used by governments from Brazil to the UAE.1:12 And finally, the institutions are keeping pace.1:15 The National Institutes of Health and the National Science Foundation just renewed support for AI research, funding everything from multi-use AI frameworks to cybersecurity training scholarships.1:26 The ecosystem is being built from the top down and the bottom up, all at once.1:31 Okay, let's go deeper on the two biggest signals today.1:34 First, that SSI valuation.1:36 The thirty-two billion dollar number is the headline.1:39 But the real story is Nvidia’s five billion dollar investment.1:43 That money isn't just cash.1:45 It’s a key.1:45 It gives Ilya Sutskever’s secretive team access to Nvidia’s brand new Vera Rubin platform, which reportedly increases computing capacity TENFOLD.1:54 This is a strategic bet by Nvidia.1:56 They are backing a pure research play on AGI, completely outside the existing orbit of OpenAI, Google, and Microsoft.2:03 Sutskever’s quote says it all: "We have research that is worthy of scaling up." He’s not talking about a product.2:10 He’s talking about a fundamental attempt to build safe superintelligence.2:14 So here’s the signal behind the noise.2:17 That valuation isn't based on revenue or customers.2:20 It’s a pure, unadulterated bet on a research outcome.2:23 It's what funding for the Manhattan Project might have looked like if it were run by venture capitalists.2:29 The speculation is obvious, and a thirty-two billion dollar company with no product is a fragile thing.2:35 But the move shows that the most important player in AI hardware believes the race for AGI requires a dedicated, massively-scaled, and secretive effort.2:44 This is the new shape of frontier research.2:47 Now for the second signal: Anthropic's Claude Mythos.2:50 This is the capability story that complements the capital story.2:54 Within sixty hours, and for about one hundred thousand dollars in API costs, the model discovered novel attacks on cryptography.3:02 It developed a new technique the researchers are calling the Möbius Bridge.3:06 It effectively halved the key strength of a digital signature scheme that was a candidate for NIST's post-quantum standard.3:14 Let's be very clear about what this means.3:16 This is NOT an AI simply running through calculations faster than a human.3:21 This is an AI discovering new pathways, new techniques, in a highly complex scientific field with minimal human guidance.3:28 The New York Times is right to say this will force faster iteration in security standards.3:33 The immediate risk is low.3:35 These discoveries don't break the encryption you're using today.3:39 But that misses the point.3:40 The point is the delta.3:42 Yesterday, cryptanalysis was a purely human endeavor.3:45 Today, an AI is a contributing researcher.3:47 It signals a new era where the security of our entire digital infrastructure will depend on an ongoing, high-speed arms race between AI-driven code-breakers and human, or AI-assisted, code-makers.3:59 So you have these two forces acting at once.4:02 On one hand, unprecedented capital is flowing into black-box AGI research.4:06 On the other, existing models are already demonstrating superhuman capabilities that change the fundamentals of digital security.4:14 The race for capability is now happening at the speed of capital, not just the speed of code.4:20 And the safety nets are being tested in real time, whether we are ready or not.
