The Arrival of the AI Singularity | 19th Oct 2026

SPEAKER_1 (0:00)

Welcome to Colaberry AI Podcast, brought to you by Colaberry AI Research Labs and Carl Foundation.

SPEAKER_2 (0:05)

Thanks for having me back. Yeah. We have a lot to cover today.

SPEAKER_1 (0:08)

We really do. So I want you to imagine handing in your doctoral thesis, like your absolute life's work.

SPEAKER_2 (0:15)

Okay, setting the stage.

SPEAKER_1 (0:16)

Right. And the automated spell checker on your laptop, it doesn't just fix a typo. It actually proves that the professor's answer key is fundamentally mathematically impossible.

SPEAKER_2 (0:27)

Oh, wow. That is quite the scenario.

SPEAKER_1 (0:29)

Yeah. And the crazy part is that isn't some science fiction thought experiment that actually just happened in the real world of physics research.

SPEAKER_2 (0:37)

It really did.

SPEAKER_1 (0:38)

So for today's deep dive, we are totally bypassing the endless internet hype cycle. We're going to look strictly at empirical methods and quantitative results.

SPEAKER_2 (0:47)

Which is so necessary right now.

SPEAKER_1 (0:48)

Exactly. Because recently, Anthropic co-founder Jack Clark delivered the speech at the International Monetary Fund. He's standing right in front of senior central bankers, and he states that the AI Singularity is no longer this, you know, theoretical problem for the year 2040

SPEAKER_2 (1:04)

He essentially said it's already arrived.

SPEAKER_1 (1:06)

He did. And to really understand that, we are pairing his speech with Anthropic's highly technical economic report. It's titled Economic Scenarios for Transformative AI.

SPEAKER_2 (1:15)

Right. And the headline claim in that report is just staggering. They're projecting that artificial intelligence could push the US gross domestic product to $44.4 trillion by 2030

SPEAKER_1 (1:26)

$44.4 trillion. I mean, just to put that into perspective, your typical macroeconomic model, they're usually looking for fractions of a percentage point.

SPEAKER_2 (1:36)

Exactly. A multi-trillion-dollar delta is unheard of.

It means we really have to look under the hood at the technical benchmarks, the specific economic modeling methods, and what the data set actually projects for the labor market.

SPEAKER_1 (1:50)

Okay, let's unpack this because dropping a 44 trillion-dollar number at the IMF, well, that requires absolute empirical receipts.

SPEAKER_2 (1:58)

Right. You can't just throw that out there without the data.

SPEAKER_1 (2:01)

Exactly. Before we can even touch a macroeconomic model, we have to establish why researchers actually believe this capability threshold has been crossed, and the benchmark data coming out over the last few months is just staggering.

SPEAKER_2 (2:14)

It really is. I mean, starting with something called humanities last exam.

SPEAKER_1 (2:18)

Yeah. Humanities last exam, which is a notoriously brutal evaluation data set, right?

SPEAKER_2 (2:24)

Oh, absolutely. It's designed to be punishing. It covers the absolute highest difficulty questions across advanced computer science, chemistry, abstract algebra, even classical literature.

SPEAKER_1 (2:36)

Basically everything.

SPEAKER_2 (2:37)

Yeah. In January of 2025, the most capable frontier model at the time scored a dismal 8%.

SPEAKER_1 (2:45)

Wow. Just 8%.

SPEAKER_2 (2:46)

Yeah. It was failing spectacularly. But then, the latest evaluation of Claude Opus 5.5 just scored 67.7%.

SPEAKER_1 (2:55)

Wait. Going from 8% to almost 68% in under two years on PhD level synthesis.

SPEAKER_2 (3:02)

Yeah. It's a massive quantitative leap.

SPEAKER_1 (3:04)

That is wild. But, you know, it's the specialized domain evaluations that really caught my attention in these sources, like the frontier math data set.

SPEAKER_2 (3:12)

Oh, right. The one compiled by hundreds of professional mathematicians.

SPEAKER_1 (3:16)

Exactly. So at the end of 2024, models had like a 2% success rate on that.

Today, they are hitting 94%.

SPEAKER_2 (3:25)

It's incredible.

SPEAKER_1 (3:26)

And the source material specifically mentions AI providing core breakthrough insights on historically unsolvable mathematics, specifically the Navier-Stokes equations and the Jacobian conjecture.

SPEAKER_2 (3:38)

Yeah. Those aren't just puzzles.

SPEAKER_1 (3:40)

Those are foundational mathematical roadblocks. I mean, the Navier-Stokes equations define fluid dynamics.

SPEAKER_2 (3:46)

Right. Like how liquids and gasses flow.

SPEAKER_1 (3:48)

Exactly.

SPEAKER_2 (3:49)

We need those equations to perfectly model everything from, say, global weather patterns to the aerodynamics of a hypersonic jet. But humans just haven't been able to fully solve the math for over a century.

SPEAKER_1 (4:01)

And then there's the Jacobian conjecture.

SPEAKER_2 (4:03)

Right. Which involves multivariable calculus mapping, dealing with these incredibly complex polynomial equations. Now, to be clear, the models aren't printing out a finished peer-reviewed paper just yet.

SPEAKER_1 (4:16)

Okay. So they aren't quite there.

SPEAKER_2 (4:17)

Not entirely. But they are generating the underlying mathematical architectures that have basically eluded the smartest humans on earth.

SPEAKER_1 (4:25)

Yeah. And that brings us to the Theophys evaluation, which is actually where my doctoral thesis analogy comes from.

SPEAKER_2 (4:32)

Oh, yes. The theoretical physics test.

SPEAKER_1 (4:35)

So the evaluators gathered 50 front-tier theoretical physics questions, and these were written by top-tier physicists using their own unpublished cutting-edge research.